You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/db/db_impl.cc

4784 lines
168 KiB

// Copyright (c) 2013, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "db/db_impl.h"
#ifndef __STDC_FORMAT_MACROS
#define __STDC_FORMAT_MACROS
#endif
#include <inttypes.h>
#include <algorithm>
#include <climits>
#include <cstdio>
#include <set>
#include <stdexcept>
#include <stdint.h>
#include <string>
#include <unordered_set>
#include <unordered_map>
#include <utility>
#include <vector>
#include "db/builder.h"
#include "db/flush_job.h"
#include "db/db_iter.h"
#include "db/dbformat.h"
#include "db/filename.h"
#include "db/log_reader.h"
#include "db/log_writer.h"
#include "db/memtable.h"
#include "db/memtable_list.h"
#include "db/merge_context.h"
#include "db/merge_helper.h"
#include "db/table_cache.h"
#include "db/table_properties_collector.h"
#include "db/forward_iterator.h"
#include "db/transaction_log_impl.h"
#include "db/version_set.h"
#include "db/write_batch_internal.h"
#include "port/port.h"
#include "rocksdb/cache.h"
#include "port/likely.h"
#include "rocksdb/compaction_filter.h"
#include "rocksdb/db.h"
#include "rocksdb/env.h"
#include "rocksdb/merge_operator.h"
#include "rocksdb/version.h"
#include "rocksdb/statistics.h"
#include "rocksdb/status.h"
#include "rocksdb/table.h"
#include "table/block.h"
#include "table/block_based_table_factory.h"
#include "table/merger.h"
#include "table/table_builder.h"
#include "table/two_level_iterator.h"
#include "util/auto_roll_logger.h"
#include "util/autovector.h"
#include "util/build_version.h"
#include "util/coding.h"
#include "util/hash_skiplist_rep.h"
#include "util/hash_linklist_rep.h"
#include "util/logging.h"
#include "util/log_buffer.h"
#include "util/mutexlock.h"
#include "util/perf_context_imp.h"
#include "util/iostats_context_imp.h"
#include "util/stop_watch.h"
#include "util/sync_point.h"
namespace rocksdb {
const std::string kDefaultColumnFamilyName("default");
void DumpRocksDBBuildVersion(Logger * log);
struct DBImpl::WriteContext {
autovector<SuperVersion*> superversions_to_free_;
autovector<log::Writer*> logs_to_free_;
~WriteContext() {
for (auto& sv : superversions_to_free_) {
delete sv;
}
for (auto& log : logs_to_free_) {
delete log;
}
}
};
struct DBImpl::CompactionState {
Compaction* const compaction;
// If there were two snapshots with seq numbers s1 and
// s2 and s1 < s2, and if we find two instances of a key k1 then lies
// entirely within s1 and s2, then the earlier version of k1 can be safely
// deleted because that version is not visible in any snapshot.
std::vector<SequenceNumber> existing_snapshots;
// Files produced by compaction
struct Output {
uint64_t number;
uint32_t path_id;
uint64_t file_size;
InternalKey smallest, largest;
SequenceNumber smallest_seqno, largest_seqno;
};
std::vector<Output> outputs;
std::list<uint64_t> allocated_file_numbers;
// State kept for output being generated
unique_ptr<WritableFile> outfile;
unique_ptr<TableBuilder> builder;
uint64_t total_bytes;
Output* current_output() { return &outputs[outputs.size()-1]; }
explicit CompactionState(Compaction* c)
: compaction(c),
total_bytes(0) {
}
// Create a client visible context of this compaction
CompactionFilter::Context GetFilterContextV1() {
CompactionFilter::Context context;
context.is_full_compaction = compaction->IsFullCompaction();
context.is_manual_compaction = compaction->IsManualCompaction();
return context;
}
// Create a client visible context of this compaction
CompactionFilterContext GetFilterContext() {
CompactionFilterContext context;
context.is_full_compaction = compaction->IsFullCompaction();
context.is_manual_compaction = compaction->IsManualCompaction();
return context;
}
std::vector<std::string> key_str_buf_;
std::vector<std::string> existing_value_str_buf_;
// new_value_buf_ will only be appended if a value changes
std::vector<std::string> new_value_buf_;
// if values_changed_buf_[i] is true
// new_value_buf_ will add a new entry with the changed value
std::vector<bool> value_changed_buf_;
// to_delete_buf_[i] is true iff key_buf_[i] is deleted
std::vector<bool> to_delete_buf_;
std::vector<std::string> other_key_str_buf_;
std::vector<std::string> other_value_str_buf_;
std::vector<Slice> combined_key_buf_;
std::vector<Slice> combined_value_buf_;
std::string cur_prefix_;
// Buffers the kv-pair that will be run through compaction filter V2
// in the future.
void BufferKeyValueSlices(const Slice& key, const Slice& value) {
key_str_buf_.emplace_back(key.ToString());
existing_value_str_buf_.emplace_back(value.ToString());
}
// Buffers the kv-pair that will not be run through compaction filter V2
// in the future.
void BufferOtherKeyValueSlices(const Slice& key, const Slice& value) {
other_key_str_buf_.emplace_back(key.ToString());
other_value_str_buf_.emplace_back(value.ToString());
}
// Add a kv-pair to the combined buffer
void AddToCombinedKeyValueSlices(const Slice& key, const Slice& value) {
// The real strings are stored in the batch buffers
combined_key_buf_.emplace_back(key);
combined_value_buf_.emplace_back(value);
}
// Merging the two buffers
void MergeKeyValueSliceBuffer(const InternalKeyComparator* comparator) {
size_t i = 0;
size_t j = 0;
size_t total_size = key_str_buf_.size() + other_key_str_buf_.size();
combined_key_buf_.reserve(total_size);
combined_value_buf_.reserve(total_size);
while (i + j < total_size) {
int comp_res = 0;
if (i < key_str_buf_.size() && j < other_key_str_buf_.size()) {
comp_res = comparator->Compare(key_str_buf_[i], other_key_str_buf_[j]);
} else if (i >= key_str_buf_.size() && j < other_key_str_buf_.size()) {
comp_res = 1;
} else if (j >= other_key_str_buf_.size() && i < key_str_buf_.size()) {
comp_res = -1;
}
if (comp_res > 0) {
AddToCombinedKeyValueSlices(other_key_str_buf_[j], other_value_str_buf_[j]);
j++;
} else if (comp_res < 0) {
AddToCombinedKeyValueSlices(key_str_buf_[i], existing_value_str_buf_[i]);
i++;
}
}
}
void CleanupBatchBuffer() {
to_delete_buf_.clear();
key_str_buf_.clear();
existing_value_str_buf_.clear();
new_value_buf_.clear();
value_changed_buf_.clear();
to_delete_buf_.shrink_to_fit();
key_str_buf_.shrink_to_fit();
existing_value_str_buf_.shrink_to_fit();
new_value_buf_.shrink_to_fit();
value_changed_buf_.shrink_to_fit();
other_key_str_buf_.clear();
other_value_str_buf_.clear();
other_key_str_buf_.shrink_to_fit();
other_value_str_buf_.shrink_to_fit();
}
void CleanupMergedBuffer() {
combined_key_buf_.clear();
combined_value_buf_.clear();
combined_key_buf_.shrink_to_fit();
combined_value_buf_.shrink_to_fit();
}
};
Options SanitizeOptions(const std::string& dbname,
const InternalKeyComparator* icmp,
const Options& src) {
auto db_options = SanitizeOptions(dbname, DBOptions(src));
auto cf_options = SanitizeOptions(icmp, ColumnFamilyOptions(src));
return Options(db_options, cf_options);
}
DBOptions SanitizeOptions(const std::string& dbname, const DBOptions& src) {
DBOptions result = src;
// result.max_open_files means an "infinite" open files.
if (result.max_open_files != -1) {
ClipToRange(&result.max_open_files, 20, 1000000);
}
if (result.info_log == nullptr) {
Status s = CreateLoggerFromOptions(dbname, result.db_log_dir, src.env,
result, &result.info_log);
if (!s.ok()) {
// No place suitable for logging
result.info_log = nullptr;
}
}
if (!result.rate_limiter) {
if (result.bytes_per_sync == 0) {
result.bytes_per_sync = 1024 * 1024;
}
}
if (result.wal_dir.empty()) {
// Use dbname as default
result.wal_dir = dbname;
}
if (result.wal_dir.back() == '/') {
result.wal_dir = result.wal_dir.substr(0, result.wal_dir.size() - 1);
}
if (result.db_paths.size() == 0) {
result.db_paths.emplace_back(dbname, std::numeric_limits<uint64_t>::max());
}
return result;
}
namespace {
Status SanitizeOptionsByTable(
const DBOptions& db_opts,
const std::vector<ColumnFamilyDescriptor>& column_families) {
Status s;
for (auto cf : column_families) {
s = cf.options.table_factory->SanitizeOptions(db_opts, cf.options);
if (!s.ok()) {
return s;
}
}
return Status::OK();
}
CompressionType GetCompressionFlush(const ImmutableCFOptions& ioptions) {
// Compressing memtable flushes might not help unless the sequential load
// optimization is used for leveled compaction. Otherwise the CPU and
// latency overhead is not offset by saving much space.
bool can_compress;
if (ioptions.compaction_style == kCompactionStyleUniversal) {
can_compress =
(ioptions.compaction_options_universal.compression_size_percent < 0);
} else {
// For leveled compress when min_level_to_compress == 0.
can_compress = ioptions.compression_per_level.empty() ||
ioptions.compression_per_level[0] != kNoCompression;
}
if (can_compress) {
return ioptions.compression;
} else {
return kNoCompression;
}
}
} // namespace
DBImpl::DBImpl(const DBOptions& options, const std::string& dbname)
: env_(options.env),
dbname_(dbname),
db_options_(SanitizeOptions(dbname, options)),
stats_(db_options_.statistics.get()),
db_lock_(nullptr),
mutex_(options.use_adaptive_mutex),
shutting_down_(false),
bg_cv_(&mutex_),
logfile_number_(0),
log_empty_(true),
default_cf_handle_(nullptr),
total_log_size_(0),
max_total_in_memory_state_(0),
tmp_batch_(),
bg_schedule_needed_(false),
bg_compaction_scheduled_(0),
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
bg_manual_only_(0),
bg_flush_scheduled_(0),
manual_compaction_(nullptr),
disable_delete_obsolete_files_(0),
delete_obsolete_files_last_run_(options.env->NowMicros()),
purge_wal_files_last_run_(0),
last_stats_dump_time_microsec_(0),
default_interval_to_delete_obsolete_WAL_(600),
flush_on_destroy_(false),
env_options_(options),
bg_work_gate_closed_(false),
refitting_level_(false),
opened_successfully_(false) {
env_->GetAbsolutePath(dbname, &db_absolute_path_);
// Reserve ten files or so for other uses and give the rest to TableCache.
// Give a large number for setting of "infinite" open files.
const int table_cache_size = (db_options_.max_open_files == -1) ?
4194304 : db_options_.max_open_files - 10;
// Reserve ten files or so for other uses and give the rest to TableCache.
table_cache_ =
NewLRUCache(table_cache_size, db_options_.table_cache_numshardbits,
db_options_.table_cache_remove_scan_count_limit);
versions_.reset(new VersionSet(dbname_, &db_options_, env_options_,
table_cache_.get(), &write_controller_));
column_family_memtables_.reset(new ColumnFamilyMemTablesImpl(
versions_->GetColumnFamilySet(), &flush_scheduler_));
DumpRocksDBBuildVersion(db_options_.info_log.get());
DumpDBFileSummary(db_options_, dbname_);
db_options_.Dump(db_options_.info_log.get());
LogFlush(db_options_.info_log);
}
DBImpl::~DBImpl() {
mutex_.Lock();
if (flush_on_destroy_) {
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (!cfd->mem()->IsEmpty()) {
cfd->Ref();
mutex_.Unlock();
FlushMemTable(cfd, FlushOptions());
mutex_.Lock();
cfd->Unref();
}
}
versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
}
// Wait for background work to finish
shutting_down_.store(true, std::memory_order_release);
while (bg_compaction_scheduled_ || bg_flush_scheduled_) {
bg_cv_.Wait();
}
flush_scheduler_.Clear();
if (default_cf_handle_ != nullptr) {
// we need to delete handle outside of lock because it does its own locking
mutex_.Unlock();
delete default_cf_handle_;
mutex_.Lock();
}
// Clean up obsolete files due to SuperVersion release.
// (1) Need to delete to obsolete files before closing because RepairDB()
// scans all existing files in the file system and builds manifest file.
// Keeping obsolete files confuses the repair process.
// (2) Need to check if we Open()/Recover() the DB successfully before
// deleting because if VersionSet recover fails (may be due to corrupted
// manifest file), it is not able to identify live files correctly. As a
// result, all "live" files can get deleted by accident. However, corrupted
// manifest is recoverable by RepairDB().
if (opened_successfully_) {
JobContext job_context;
FindObsoleteFiles(&job_context, true);
// manifest number starting from 2
job_context.manifest_file_number = 1;
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
}
// versions need to be destroyed before table_cache since it can hold
// references to table_cache.
versions_.reset();
mutex_.Unlock();
if (db_lock_ != nullptr) {
env_->UnlockFile(db_lock_);
}
LogFlush(db_options_.info_log);
}
Status DBImpl::NewDB() {
VersionEdit new_db;
new_db.SetLogNumber(0);
new_db.SetNextFile(2);
new_db.SetLastSequence(0);
Log(db_options_.info_log, "Creating manifest 1 \n");
const std::string manifest = DescriptorFileName(dbname_, 1);
unique_ptr<WritableFile> file;
Status s = env_->NewWritableFile(
manifest, &file, env_->OptimizeForManifestWrite(env_options_));
if (!s.ok()) {
return s;
}
file->SetPreallocationBlockSize(db_options_.manifest_preallocation_size);
{
log::Writer log(std::move(file));
std::string record;
new_db.EncodeTo(&record);
s = log.AddRecord(record);
}
if (s.ok()) {
// Make "CURRENT" file that points to the new manifest file.
s = SetCurrentFile(env_, dbname_, 1, db_directory_.get());
} else {
env_->DeleteFile(manifest);
}
return s;
}
void DBImpl::MaybeIgnoreError(Status* s) const {
if (s->ok() || db_options_.paranoid_checks) {
// No change needed
} else {
Log(db_options_.info_log, "Ignoring error %s", s->ToString().c_str());
*s = Status::OK();
}
}
const Status DBImpl::CreateArchivalDirectory() {
if (db_options_.WAL_ttl_seconds > 0 || db_options_.WAL_size_limit_MB > 0) {
std::string archivalPath = ArchivalDirectory(db_options_.wal_dir);
return env_->CreateDirIfMissing(archivalPath);
}
return Status::OK();
}
void DBImpl::PrintStatistics() {
auto dbstats = db_options_.statistics.get();
if (dbstats) {
Log(db_options_.info_log,
"STATISTCS:\n %s",
dbstats->ToString().c_str());
}
}
void DBImpl::MaybeDumpStats() {
if (db_options_.stats_dump_period_sec == 0) return;
const uint64_t now_micros = env_->NowMicros();
if (last_stats_dump_time_microsec_ +
db_options_.stats_dump_period_sec * 1000000
<= now_micros) {
// Multiple threads could race in here simultaneously.
// However, the last one will update last_stats_dump_time_microsec_
// atomically. We could see more than one dump during one dump
// period in rare cases.
last_stats_dump_time_microsec_ = now_micros;
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
bool tmp1 = false;
bool tmp2 = false;
DBPropertyType cf_property_type =
GetPropertyType("rocksdb.cfstats", &tmp1, &tmp2);
DBPropertyType db_property_type =
GetPropertyType("rocksdb.dbstats", &tmp1, &tmp2);
std::string stats;
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
{
MutexLock l(&mutex_);
for (auto cfd : *versions_->GetColumnFamilySet()) {
cfd->internal_stats()->GetStringProperty(cf_property_type,
"rocksdb.cfstats", &stats);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
}
default_cf_internal_stats_->GetStringProperty(db_property_type,
"rocksdb.dbstats", &stats);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
}
Log(db_options_.info_log, "------- DUMPING STATS -------");
Log(db_options_.info_log, "%s", stats.c_str());
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
PrintStatistics();
}
}
// Returns the list of live files in 'sst_live' and the list
// of all files in the filesystem in 'candidate_files'.
// no_full_scan = true -- never do the full scan using GetChildren()
// force = false -- don't force the full scan, except every
// db_options_.delete_obsolete_files_period_micros
// force = true -- force the full scan
void DBImpl::FindObsoleteFiles(JobContext* job_context, bool force,
bool no_full_scan) {
mutex_.AssertHeld();
// if deletion is disabled, do nothing
if (disable_delete_obsolete_files_ > 0) {
return;
}
bool doing_the_full_scan = false;
// logic for figurint out if we're doing the full scan
if (no_full_scan) {
doing_the_full_scan = false;
} else if (force || db_options_.delete_obsolete_files_period_micros == 0) {
doing_the_full_scan = true;
} else {
const uint64_t now_micros = env_->NowMicros();
if (delete_obsolete_files_last_run_ +
db_options_.delete_obsolete_files_period_micros < now_micros) {
doing_the_full_scan = true;
delete_obsolete_files_last_run_ = now_micros;
}
}
// get obsolete files
versions_->GetObsoleteFiles(&job_context->sst_delete_files);
// store the current filenum, lognum, etc
job_context->manifest_file_number = versions_->ManifestFileNumber();
job_context->pending_manifest_file_number =
versions_->PendingManifestFileNumber();
job_context->log_number = versions_->MinLogNumber();
job_context->prev_log_number = versions_->PrevLogNumber();
if (!doing_the_full_scan && !job_context->HaveSomethingToDelete()) {
// avoid filling up sst_live if we're sure that we
// are not going to do the full scan and that we don't have
// anything to delete at the moment
return;
}
// don't delete live files
for (auto pair : pending_outputs_) {
job_context->sst_live.emplace_back(pair.first, pair.second, 0);
}
versions_->AddLiveFiles(&job_context->sst_live);
if (doing_the_full_scan) {
for (uint32_t path_id = 0;
path_id < db_options_.db_paths.size(); path_id++) {
// set of all files in the directory. We'll exclude files that are still
// alive in the subsequent processings.
std::vector<std::string> files;
env_->GetChildren(db_options_.db_paths[path_id].path,
&files); // Ignore errors
for (std::string file : files) {
// TODO(icanadi) clean up this mess to avoid having one-off "/" prefixes
job_context->candidate_files.emplace_back("/" + file, path_id);
}
}
//Add log files in wal_dir
if (db_options_.wal_dir != dbname_) {
std::vector<std::string> log_files;
env_->GetChildren(db_options_.wal_dir, &log_files); // Ignore errors
for (std::string log_file : log_files) {
job_context->candidate_files.emplace_back(log_file, 0);
}
}
// Add info log files in db_log_dir
if (!db_options_.db_log_dir.empty() && db_options_.db_log_dir != dbname_) {
std::vector<std::string> info_log_files;
// Ignore errors
env_->GetChildren(db_options_.db_log_dir, &info_log_files);
for (std::string log_file : info_log_files) {
job_context->candidate_files.emplace_back(log_file, 0);
}
}
}
}
namespace {
bool CompareCandidateFile(const JobContext::CandidateFileInfo& first,
const JobContext::CandidateFileInfo& second) {
if (first.file_name > second.file_name) {
return true;
} else if (first.file_name < second.file_name) {
return false;
} else {
return (first.path_id > second.path_id);
}
}
}; // namespace
// Diffs the files listed in filenames and those that do not
// belong to live files are posibly removed. Also, removes all the
// files in sst_delete_files and log_delete_files.
// It is not necessary to hold the mutex when invoking this method.
void DBImpl::PurgeObsoleteFiles(const JobContext& state) {
// we'd better have sth to delete
assert(state.HaveSomethingToDelete());
// this checks if FindObsoleteFiles() was run before. If not, don't do
// PurgeObsoleteFiles(). If FindObsoleteFiles() was run, we need to also
// run PurgeObsoleteFiles(), even if disable_delete_obsolete_files_ is true
if (state.manifest_file_number == 0) {
return;
}
// Now, convert live list to an unordered map, WITHOUT mutex held;
// set is slow.
std::unordered_map<uint64_t, const FileDescriptor*> sst_live_map;
for (const FileDescriptor& fd : state.sst_live) {
sst_live_map[fd.GetNumber()] = &fd;
}
auto candidate_files = state.candidate_files;
candidate_files.reserve(candidate_files.size() +
state.sst_delete_files.size() +
state.log_delete_files.size());
// We may ignore the dbname when generating the file names.
const char* kDumbDbName = "";
for (auto file : state.sst_delete_files) {
candidate_files.emplace_back(
MakeTableFileName(kDumbDbName, file->fd.GetNumber()),
file->fd.GetPathId());
delete file;
}
for (auto file_num : state.log_delete_files) {
if (file_num > 0) {
candidate_files.emplace_back(LogFileName(kDumbDbName, file_num).substr(1),
0);
}
}
// dedup state.candidate_files so we don't try to delete the same
// file twice
sort(candidate_files.begin(), candidate_files.end(), CompareCandidateFile);
candidate_files.erase(unique(candidate_files.begin(), candidate_files.end()),
candidate_files.end());
std::vector<std::string> old_info_log_files;
InfoLogPrefix info_log_prefix(!db_options_.db_log_dir.empty(), dbname_);
for (const auto& candidate_file : candidate_files) {
std::string to_delete = candidate_file.file_name;
uint32_t path_id = candidate_file.path_id;
uint64_t number;
FileType type;
// Ignore file if we cannot recognize it.
if (!ParseFileName(to_delete, &number, info_log_prefix.prefix, &type)) {
continue;
}
bool keep = true;
switch (type) {
case kLogFile:
keep = ((number >= state.log_number) ||
(number == state.prev_log_number));
break;
case kDescriptorFile:
// Keep my manifest file, and any newer incarnations'
// (can happen during manifest roll)
keep = (number >= state.manifest_file_number);
break;
case kTableFile:
keep = (sst_live_map.find(number) != sst_live_map.end());
break;
case kTempFile:
// Any temp files that are currently being written to must
// be recorded in pending_outputs_, which is inserted into "live".
// Also, SetCurrentFile creates a temp file when writing out new
// manifest, which is equal to state.pending_manifest_file_number. We
// should not delete that file
keep = (sst_live_map.find(number) != sst_live_map.end()) ||
(number == state.pending_manifest_file_number);
break;
case kInfoLogFile:
keep = true;
if (number != 0) {
old_info_log_files.push_back(to_delete);
}
break;
case kCurrentFile:
case kDBLockFile:
case kIdentityFile:
case kMetaDatabase:
keep = true;
break;
}
if (keep) {
continue;
}
std::string fname;
if (type == kTableFile) {
// evict from cache
TableCache::Evict(table_cache_.get(), number);
fname = TableFileName(db_options_.db_paths, number, path_id);
} else {
fname = ((type == kLogFile) ?
db_options_.wal_dir : dbname_) + "/" + to_delete;
}
if (type == kLogFile &&
(db_options_.WAL_ttl_seconds > 0 ||
db_options_.WAL_size_limit_MB > 0)) {
auto archived_log_name = ArchivedLogFileName(db_options_.wal_dir, number);
// The sync point below is used in (DBTest,TransactionLogIteratorRace)
TEST_SYNC_POINT("DBImpl::PurgeObsoleteFiles:1");
Status s = env_->RenameFile(fname, archived_log_name);
// The sync point below is used in (DBTest,TransactionLogIteratorRace)
TEST_SYNC_POINT("DBImpl::PurgeObsoleteFiles:2");
Log(db_options_.info_log,
"Move log file %s to %s -- %s\n",
fname.c_str(), archived_log_name.c_str(), s.ToString().c_str());
} else {
Status s = env_->DeleteFile(fname);
Log(db_options_.info_log, "Delete %s type=%d #%" PRIu64 " -- %s\n",
fname.c_str(), type, number, s.ToString().c_str());
}
}
// Delete old info log files.
size_t old_info_log_file_count = old_info_log_files.size();
if (old_info_log_file_count >= db_options_.keep_log_file_num) {
std::sort(old_info_log_files.begin(), old_info_log_files.end());
size_t end = old_info_log_file_count - db_options_.keep_log_file_num;
for (unsigned int i = 0; i <= end; i++) {
std::string& to_delete = old_info_log_files.at(i);
std::string full_path_to_delete = (db_options_.db_log_dir.empty() ?
dbname_ : db_options_.db_log_dir) + "/" + to_delete;
Log(db_options_.info_log, "Delete info log file %s\n",
full_path_to_delete.c_str());
Status s = env_->DeleteFile(full_path_to_delete);
if (!s.ok()) {
Log(db_options_.info_log, "Delete info log file %s FAILED -- %s\n",
to_delete.c_str(), s.ToString().c_str());
}
}
}
PurgeObsoleteWALFiles();
LogFlush(db_options_.info_log);
}
void DBImpl::DeleteObsoleteFiles() {
mutex_.AssertHeld();
JobContext job_context;
FindObsoleteFiles(&job_context, true);
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
}
#ifndef ROCKSDB_LITE
// 1. Go through all archived files and
// a. if ttl is enabled, delete outdated files
// b. if archive size limit is enabled, delete empty files,
// compute file number and size.
// 2. If size limit is enabled:
// a. compute how many files should be deleted
// b. get sorted non-empty archived logs
// c. delete what should be deleted
void DBImpl::PurgeObsoleteWALFiles() {
bool const ttl_enabled = db_options_.WAL_ttl_seconds > 0;
bool const size_limit_enabled = db_options_.WAL_size_limit_MB > 0;
if (!ttl_enabled && !size_limit_enabled) {
return;
}
int64_t current_time;
Status s = env_->GetCurrentTime(&current_time);
if (!s.ok()) {
Log(db_options_.info_log, "Can't get current time: %s",
s.ToString().c_str());
assert(false);
return;
}
uint64_t const now_seconds = static_cast<uint64_t>(current_time);
uint64_t const time_to_check = (ttl_enabled && !size_limit_enabled) ?
db_options_.WAL_ttl_seconds / 2 : default_interval_to_delete_obsolete_WAL_;
if (purge_wal_files_last_run_ + time_to_check > now_seconds) {
return;
}
purge_wal_files_last_run_ = now_seconds;
std::string archival_dir = ArchivalDirectory(db_options_.wal_dir);
std::vector<std::string> files;
s = env_->GetChildren(archival_dir, &files);
if (!s.ok()) {
Log(db_options_.info_log, "Can't get archive files: %s",
s.ToString().c_str());
assert(false);
return;
}
size_t log_files_num = 0;
uint64_t log_file_size = 0;
for (auto& f : files) {
uint64_t number;
FileType type;
if (ParseFileName(f, &number, &type) && type == kLogFile) {
std::string const file_path = archival_dir + "/" + f;
if (ttl_enabled) {
uint64_t file_m_time;
Status const s = env_->GetFileModificationTime(file_path,
&file_m_time);
if (!s.ok()) {
Log(db_options_.info_log, "Can't get file mod time: %s: %s",
file_path.c_str(), s.ToString().c_str());
continue;
}
if (now_seconds - file_m_time > db_options_.WAL_ttl_seconds) {
Status const s = env_->DeleteFile(file_path);
if (!s.ok()) {
Log(db_options_.info_log, "Can't delete file: %s: %s",
file_path.c_str(), s.ToString().c_str());
continue;
} else {
MutexLock l(&read_first_record_cache_mutex_);
read_first_record_cache_.erase(number);
}
continue;
}
}
if (size_limit_enabled) {
uint64_t file_size;
Status const s = env_->GetFileSize(file_path, &file_size);
if (!s.ok()) {
Log(db_options_.info_log, "Can't get file size: %s: %s",
file_path.c_str(), s.ToString().c_str());
return;
} else {
if (file_size > 0) {
log_file_size = std::max(log_file_size, file_size);
++log_files_num;
} else {
Status s = env_->DeleteFile(file_path);
if (!s.ok()) {
Log(db_options_.info_log, "Can't delete file: %s: %s",
file_path.c_str(), s.ToString().c_str());
continue;
} else {
MutexLock l(&read_first_record_cache_mutex_);
read_first_record_cache_.erase(number);
}
}
}
}
}
}
if (0 == log_files_num || !size_limit_enabled) {
return;
}
size_t const files_keep_num = db_options_.WAL_size_limit_MB *
1024 * 1024 / log_file_size;
if (log_files_num <= files_keep_num) {
return;
}
size_t files_del_num = log_files_num - files_keep_num;
VectorLogPtr archived_logs;
GetSortedWalsOfType(archival_dir, archived_logs, kArchivedLogFile);
if (files_del_num > archived_logs.size()) {
Log(db_options_.info_log, "Trying to delete more archived log files than "
"exist. Deleting all");
files_del_num = archived_logs.size();
}
for (size_t i = 0; i < files_del_num; ++i) {
std::string const file_path = archived_logs[i]->PathName();
Status const s = DeleteFile(file_path);
if (!s.ok()) {
Log(db_options_.info_log, "Can't delete file: %s: %s",
file_path.c_str(), s.ToString().c_str());
continue;
} else {
MutexLock l(&read_first_record_cache_mutex_);
read_first_record_cache_.erase(archived_logs[i]->LogNumber());
}
}
}
namespace {
struct CompareLogByPointer {
bool operator()(const unique_ptr<LogFile>& a, const unique_ptr<LogFile>& b) {
LogFileImpl* a_impl = dynamic_cast<LogFileImpl*>(a.get());
LogFileImpl* b_impl = dynamic_cast<LogFileImpl*>(b.get());
return *a_impl < *b_impl;
}
};
}
Status DBImpl::GetSortedWalsOfType(const std::string& path,
VectorLogPtr& log_files,
WalFileType log_type) {
std::vector<std::string> all_files;
const Status status = env_->GetChildren(path, &all_files);
if (!status.ok()) {
return status;
}
log_files.reserve(all_files.size());
for (const auto& f : all_files) {
uint64_t number;
FileType type;
if (ParseFileName(f, &number, &type) && type == kLogFile) {
SequenceNumber sequence;
Status s = ReadFirstRecord(log_type, number, &sequence);
if (!s.ok()) {
return s;
}
if (sequence == 0) {
// empty file
continue;
}
// Reproduce the race condition where a log file is moved
// to archived dir, between these two sync points, used in
// (DBTest,TransactionLogIteratorRace)
TEST_SYNC_POINT("DBImpl::GetSortedWalsOfType:1");
TEST_SYNC_POINT("DBImpl::GetSortedWalsOfType:2");
uint64_t size_bytes;
s = env_->GetFileSize(LogFileName(path, number), &size_bytes);
// re-try in case the alive log file has been moved to archive.
if (!s.ok() && log_type == kAliveLogFile &&
env_->FileExists(ArchivedLogFileName(path, number))) {
s = env_->GetFileSize(ArchivedLogFileName(path, number), &size_bytes);
}
if (!s.ok()) {
return s;
}
log_files.push_back(std::move(unique_ptr<LogFile>(
new LogFileImpl(number, log_type, sequence, size_bytes))));
}
}
CompareLogByPointer compare_log_files;
std::sort(log_files.begin(), log_files.end(), compare_log_files);
return status;
}
Status DBImpl::RetainProbableWalFiles(VectorLogPtr& all_logs,
const SequenceNumber target) {
int64_t start = 0; // signed to avoid overflow when target is < first file.
int64_t end = static_cast<int64_t>(all_logs.size()) - 1;
// Binary Search. avoid opening all files.
while (end >= start) {
int64_t mid = start + (end - start) / 2; // Avoid overflow.
SequenceNumber current_seq_num = all_logs.at(mid)->StartSequence();
if (current_seq_num == target) {
end = mid;
break;
} else if (current_seq_num < target) {
start = mid + 1;
} else {
end = mid - 1;
}
}
// end could be -ve.
size_t start_index = std::max(static_cast<int64_t>(0), end);
// The last wal file is always included
all_logs.erase(all_logs.begin(), all_logs.begin() + start_index);
return Status::OK();
}
Status DBImpl::ReadFirstRecord(const WalFileType type, const uint64_t number,
SequenceNumber* sequence) {
if (type != kAliveLogFile && type != kArchivedLogFile) {
return Status::NotSupported("File Type Not Known " + std::to_string(type));
}
{
MutexLock l(&read_first_record_cache_mutex_);
auto itr = read_first_record_cache_.find(number);
if (itr != read_first_record_cache_.end()) {
*sequence = itr->second;
return Status::OK();
}
}
Status s;
if (type == kAliveLogFile) {
std::string fname = LogFileName(db_options_.wal_dir, number);
s = ReadFirstLine(fname, sequence);
if (env_->FileExists(fname) && !s.ok()) {
// return any error that is not caused by non-existing file
return s;
}
}
if (type == kArchivedLogFile || !s.ok()) {
// check if the file got moved to archive.
std::string archived_file =
ArchivedLogFileName(db_options_.wal_dir, number);
s = ReadFirstLine(archived_file, sequence);
}
if (s.ok() && *sequence != 0) {
MutexLock l(&read_first_record_cache_mutex_);
read_first_record_cache_.insert({number, *sequence});
}
return s;
}
// the function returns status.ok() and sequence == 0 if the file exists, but is
// empty
Status DBImpl::ReadFirstLine(const std::string& fname,
SequenceNumber* sequence) {
struct LogReporter : public log::Reader::Reporter {
Env* env;
Logger* info_log;
const char* fname;
Status* status;
bool ignore_error; // true if db_options_.paranoid_checks==false
virtual void Corruption(size_t bytes, const Status& s) {
Log(info_log, "%s%s: dropping %d bytes; %s",
(this->ignore_error ? "(ignoring error) " : ""), fname,
static_cast<int>(bytes), s.ToString().c_str());
if (this->status->ok()) {
// only keep the first error
*this->status = s;
}
}
};
unique_ptr<SequentialFile> file;
Status status = env_->NewSequentialFile(fname, &file, env_options_);
if (!status.ok()) {
return status;
}
LogReporter reporter;
reporter.env = env_;
reporter.info_log = db_options_.info_log.get();
reporter.fname = fname.c_str();
reporter.status = &status;
reporter.ignore_error = !db_options_.paranoid_checks;
log::Reader reader(std::move(file), &reporter, true /*checksum*/,
0 /*initial_offset*/);
std::string scratch;
Slice record;
if (reader.ReadRecord(&record, &scratch) &&
(status.ok() || !db_options_.paranoid_checks)) {
if (record.size() < 12) {
reporter.Corruption(record.size(),
Status::Corruption("log record too small"));
// TODO read record's till the first no corrupt entry?
} else {
WriteBatch batch;
WriteBatchInternal::SetContents(&batch, record);
*sequence = WriteBatchInternal::Sequence(&batch);
return Status::OK();
}
}
// ReadRecord returns false on EOF, which means that the log file is empty. we
// return status.ok() in that case and set sequence number to 0
*sequence = 0;
return status;
}
#endif // ROCKSDB_LITE
Status DBImpl::Recover(
const std::vector<ColumnFamilyDescriptor>& column_families, bool read_only,
bool error_if_log_file_exist) {
mutex_.AssertHeld();
bool is_new_db = false;
assert(db_lock_ == nullptr);
Refactor Recover() code Summary: This diff does two things: * Rethinks how we call Recover() with read_only option. Before, we call it with pointer to memtable where we'd like to apply those changes to. This memtable is set in db_impl_readonly.cc and it's actually DBImpl::mem_. Why don't we just apply updates to mem_ right away? It seems more intuitive. * Changes when we apply updates to manifest. Before, the process is to recover all the logs, flush it to sst files and then do one giant commit that atomically adds all recovered sst files and sets the next log number. This works good enough, but causes some small troubles for my column family approach, since I can't have one VersionEdit apply to more than single column family[1]. The change here is to commit the files recovered from logs right away. Here is the state of the world before the change: 1. Recover log 5, add new sst files to edit 2. Recover log 7, add new sst files to edit 3. Recover log 8, add new sst files to edit 4. Commit all added sst files to manifest and mark log files 5, 7 and 8 as recoverd (via SetLogNumber(9) function) After the change, we'll do: 1. Recover log 5, commit the new sst files and set log 5 as recovered 2. Recover log 7, commit the new sst files and set log 7 as recovered 3. Recover log 8, commit the new sst files and set log 8 as recovered The added (small) benefit is that if we fail after (2), the new recovery will only have to recover log 8. In previous case, we'll have to restart the recovery from the beginning. The bigger benefit will be to enable easier integration of multiple column families in Recovery code path. [1] I'm happy to dicuss this decison, but I believe this is the cleanest way to go. It also makes backward compatibility much easier. We don't have a requirement of adding multiple column families atomically. Test Plan: make check Reviewers: dhruba, haobo, kailiu, sdong Reviewed By: kailiu CC: leveldb Differential Revision: https://reviews.facebook.net/D15237
11 years ago
if (!read_only) {
// We call CreateDirIfMissing() as the directory may already exist (if we
// are reopening a DB), when this happens we don't want creating the
// directory to cause an error. However, we need to check if creating the
// directory fails or else we may get an obscure message about the lock
// file not existing. One real-world example of this occurring is if
// env->CreateDirIfMissing() doesn't create intermediate directories, e.g.
// when dbname_ is "dir/db" but when "dir" doesn't exist.
Status s = env_->CreateDirIfMissing(dbname_);
if (!s.ok()) {
return s;
}
for (auto& db_path : db_options_.db_paths) {
s = env_->CreateDirIfMissing(db_path.path);
if (!s.ok()) {
return s;
}
}
s = env_->NewDirectory(dbname_, &db_directory_);
if (!s.ok()) {
return s;
}
s = env_->LockFile(LockFileName(dbname_), &db_lock_);
if (!s.ok()) {
return s;
}
if (!env_->FileExists(CurrentFileName(dbname_))) {
if (db_options_.create_if_missing) {
s = NewDB();
is_new_db = true;
if (!s.ok()) {
return s;
}
} else {
return Status::InvalidArgument(
dbname_, "does not exist (create_if_missing is false)");
}
} else {
if (db_options_.error_if_exists) {
return Status::InvalidArgument(
dbname_, "exists (error_if_exists is true)");
}
}
Dbid feature Summary: Create a new type of file on startup if it doesn't already exist called DBID. This will store a unique number generated from boost library's uuid header file. The use-case is to identify the case of a db losing all its data and coming back up either empty or from an image(backup/live replica's recovery) the key point to note is that DBID is not stored in a backup or db snapshot It's preferable to use Boost for uuid because: 1) A non-standard way of generating uuid is not good 2) /proc/sys/kernel/random/uuid generates a uuid but only on linux environments and the solution would not be clean 3) c++ doesn't have any direct way to get a uuid 4) Boost is a very good library that was already having linkage in rocksdb from third-party Note: I had to update the TOOLCHAIN_REV in build files to get latest verison of boost from third-party as the older version had a bug. I had to put Wno-uninitialized in Makefile because boost-1.51 has an unitialized variable and rocksdb would not comiple otherwise. Latet open-source for boost is 1.54 but is not there in third-party. I have notified the concerned people in fbcode about it. @kailiu : While releasing to third-party, an additional dependency will need to be created for boost in TARGETS file. I can help identify. Test Plan: Expand db_test to test 2 cases 1) Restarting db with Id file present - verify that no change to Id 2)Restarting db with Id file deleted - verify that a different Id is there after reopen Also run make all check Reviewers: dhruba, haobo, kailiu, sdong Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13587
11 years ago
// Check for the IDENTITY file and create it if not there
if (!env_->FileExists(IdentityFileName(dbname_))) {
s = SetIdentityFile(env_, dbname_);
if (!s.ok()) {
return s;
}
}
}
Status s = versions_->Recover(column_families, read_only);
if (db_options_.paranoid_checks && s.ok()) {
s = CheckConsistency();
}
if (s.ok()) {
SequenceNumber max_sequence(0);
default_cf_handle_ = new ColumnFamilyHandleImpl(
versions_->GetColumnFamilySet()->GetDefault(), this, &mutex_);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
default_cf_internal_stats_ = default_cf_handle_->cfd()->internal_stats();
single_column_family_mode_ =
versions_->GetColumnFamilySet()->NumberOfColumnFamilies() == 1;
// Recover from all newer log files than the ones named in the
// descriptor (new log files may have been added by the previous
// incarnation without registering them in the descriptor).
//
// Note that PrevLogNumber() is no longer used, but we pay
// attention to it in case we are recovering a database
// produced by an older version of rocksdb.
const uint64_t min_log = versions_->MinLogNumber();
const uint64_t prev_log = versions_->PrevLogNumber();
std::vector<std::string> filenames;
s = env_->GetChildren(db_options_.wal_dir, &filenames);
if (!s.ok()) {
return s;
}
std::vector<uint64_t> logs;
for (size_t i = 0; i < filenames.size(); i++) {
uint64_t number;
FileType type;
if (ParseFileName(filenames[i], &number, &type) && type == kLogFile) {
if (is_new_db) {
return Status::Corruption(
"While creating a new Db, wal_dir contains "
"existing log file: ",
filenames[i]);
} else if ((number >= min_log) || (number == prev_log)) {
logs.push_back(number);
}
}
}
if (logs.size() > 0 && error_if_log_file_exist) {
return Status::Corruption(""
"The db was opened in readonly mode with error_if_log_file_exist"
"flag but a log file already exists");
}
if (!logs.empty()) {
// Recover in the order in which the logs were generated
std::sort(logs.begin(), logs.end());
s = RecoverLogFiles(logs, &max_sequence, read_only);
if (!s.ok()) {
// Clear memtables if recovery failed
for (auto cfd : *versions_->GetColumnFamilySet()) {
cfd->CreateNewMemtable(*cfd->GetLatestMutableCFOptions());
}
}
}
SetTickerCount(stats_, SEQUENCE_NUMBER, versions_->LastSequence());
}
// Initial value
max_total_in_memory_state_ = 0;
for (auto cfd : *versions_->GetColumnFamilySet()) {
auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
max_total_in_memory_state_ += mutable_cf_options->write_buffer_size *
mutable_cf_options->max_write_buffer_number;
}
return s;
}
// REQUIRES: log_numbers are sorted in ascending order
Status DBImpl::RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
SequenceNumber* max_sequence, bool read_only) {
struct LogReporter : public log::Reader::Reporter {
Env* env;
Logger* info_log;
const char* fname;
Status* status; // nullptr if db_options_.paranoid_checks==false or
// db_options_.skip_log_error_on_recovery==true
virtual void Corruption(size_t bytes, const Status& s) {
Log(info_log, "%s%s: dropping %d bytes; %s",
(this->status == nullptr ? "(ignoring error) " : ""),
fname, static_cast<int>(bytes), s.ToString().c_str());
if (this->status != nullptr && this->status->ok()) *this->status = s;
}
};
mutex_.AssertHeld();
Status status;
std::unordered_map<int, VersionEdit> version_edits;
// no need to refcount because iteration is under mutex
for (auto cfd : *versions_->GetColumnFamilySet()) {
VersionEdit edit;
edit.SetColumnFamily(cfd->GetID());
version_edits.insert({cfd->GetID(), edit});
}
Refactor Recover() code Summary: This diff does two things: * Rethinks how we call Recover() with read_only option. Before, we call it with pointer to memtable where we'd like to apply those changes to. This memtable is set in db_impl_readonly.cc and it's actually DBImpl::mem_. Why don't we just apply updates to mem_ right away? It seems more intuitive. * Changes when we apply updates to manifest. Before, the process is to recover all the logs, flush it to sst files and then do one giant commit that atomically adds all recovered sst files and sets the next log number. This works good enough, but causes some small troubles for my column family approach, since I can't have one VersionEdit apply to more than single column family[1]. The change here is to commit the files recovered from logs right away. Here is the state of the world before the change: 1. Recover log 5, add new sst files to edit 2. Recover log 7, add new sst files to edit 3. Recover log 8, add new sst files to edit 4. Commit all added sst files to manifest and mark log files 5, 7 and 8 as recoverd (via SetLogNumber(9) function) After the change, we'll do: 1. Recover log 5, commit the new sst files and set log 5 as recovered 2. Recover log 7, commit the new sst files and set log 7 as recovered 3. Recover log 8, commit the new sst files and set log 8 as recovered The added (small) benefit is that if we fail after (2), the new recovery will only have to recover log 8. In previous case, we'll have to restart the recovery from the beginning. The bigger benefit will be to enable easier integration of multiple column families in Recovery code path. [1] I'm happy to dicuss this decison, but I believe this is the cleanest way to go. It also makes backward compatibility much easier. We don't have a requirement of adding multiple column families atomically. Test Plan: make check Reviewers: dhruba, haobo, kailiu, sdong Reviewed By: kailiu CC: leveldb Differential Revision: https://reviews.facebook.net/D15237
11 years ago
for (auto log_number : log_numbers) {
// The previous incarnation may not have written any MANIFEST
// records after allocating this log number. So we manually
// update the file number allocation counter in VersionSet.
versions_->MarkFileNumberUsed(log_number);
// Open the log file
std::string fname = LogFileName(db_options_.wal_dir, log_number);
unique_ptr<SequentialFile> file;
status = env_->NewSequentialFile(fname, &file, env_options_);
if (!status.ok()) {
MaybeIgnoreError(&status);
if (!status.ok()) {
return status;
} else {
// Fail with one log file, but that's ok.
// Try next one.
continue;
}
}
// Create the log reader.
LogReporter reporter;
reporter.env = env_;
reporter.info_log = db_options_.info_log.get();
reporter.fname = fname.c_str();
reporter.status =
(db_options_.paranoid_checks && !db_options_.skip_log_error_on_recovery
? &status
: nullptr);
// We intentially make log::Reader do checksumming even if
// paranoid_checks==false so that corruptions cause entire commits
// to be skipped instead of propagating bad information (like overly
// large sequence numbers).
log::Reader reader(std::move(file), &reporter, true /*checksum*/,
0 /*initial_offset*/);
Log(db_options_.info_log, "Recovering log #%" PRIu64 "", log_number);
// Read all the records and add to a memtable
std::string scratch;
Slice record;
WriteBatch batch;
while (reader.ReadRecord(&record, &scratch)) {
if (record.size() < 12) {
reporter.Corruption(record.size(),
Status::Corruption("log record too small"));
continue;
}
WriteBatchInternal::SetContents(&batch, record);
// If column family was not found, it might mean that the WAL write
// batch references to the column family that was dropped after the
// insert. We don't want to fail the whole write batch in that case --
// we just ignore the update.
// That's why we set ignore missing column families to true
status = WriteBatchInternal::InsertInto(
&batch, column_family_memtables_.get(), true, log_number);
MaybeIgnoreError(&status);
if (!status.ok()) {
return status;
}
const SequenceNumber last_seq = WriteBatchInternal::Sequence(&batch) +
WriteBatchInternal::Count(&batch) - 1;
if (last_seq > *max_sequence) {
*max_sequence = last_seq;
}
if (!read_only) {
// we can do this because this is called before client has access to the
// DB and there is only a single thread operating on DB
ColumnFamilyData* cfd;
while ((cfd = flush_scheduler_.GetNextColumnFamily()) != nullptr) {
cfd->Unref();
// If this asserts, it means that InsertInto failed in
// filtering updates to already-flushed column families
assert(cfd->GetLogNumber() <= log_number);
auto iter = version_edits.find(cfd->GetID());
assert(iter != version_edits.end());
VersionEdit* edit = &iter->second;
status = WriteLevel0TableForRecovery(cfd, cfd->mem(), edit);
if (!status.ok()) {
// Reflect errors immediately so that conditions like full
// file-systems cause the DB::Open() to fail.
return status;
}
cfd->CreateNewMemtable(*cfd->GetLatestMutableCFOptions());
}
}
}
flush_scheduler_.Clear();
if (versions_->LastSequence() < *max_sequence) {
versions_->SetLastSequence(*max_sequence);
}
}
if (!read_only) {
// no need to refcount since client still doesn't have access
// to the DB and can not drop column families while we iterate
auto max_log_number = log_numbers.back();
for (auto cfd : *versions_->GetColumnFamilySet()) {
auto iter = version_edits.find(cfd->GetID());
assert(iter != version_edits.end());
VersionEdit* edit = &iter->second;
if (cfd->GetLogNumber() > max_log_number) {
// Column family cfd has already flushed the data
// from all logs. Memtable has to be empty because
// we filter the updates based on log_number
// (in WriteBatch::InsertInto)
assert(cfd->mem()->GetFirstSequenceNumber() == 0);
assert(edit->NumEntries() == 0);
continue;
}
// flush the final memtable (if non-empty)
if (cfd->mem()->GetFirstSequenceNumber() != 0) {
status = WriteLevel0TableForRecovery(cfd, cfd->mem(), edit);
if (!status.ok()) {
// Recovery failed
break;
}
cfd->CreateNewMemtable(*cfd->GetLatestMutableCFOptions());
}
// write MANIFEST with update
// writing log_number in the manifest means that any log file
// with number strongly less than (log_number + 1) is already
// recovered and should be ignored on next reincarnation.
// Since we already recovered max_log_number, we want all logs
// with numbers `<= max_log_number` (includes this one) to be ignored
edit->SetLogNumber(max_log_number + 1);
// we must mark the next log number as used, even though it's
// not actually used. that is because VersionSet assumes
// VersionSet::next_file_number_ always to be strictly greater than any
// log number
versions_->MarkFileNumberUsed(max_log_number + 1);
status = versions_->LogAndApply(
cfd, *cfd->GetLatestMutableCFOptions(), edit, &mutex_);
if (!status.ok()) {
// Recovery failed
break;
}
}
}
Refactor Recover() code Summary: This diff does two things: * Rethinks how we call Recover() with read_only option. Before, we call it with pointer to memtable where we'd like to apply those changes to. This memtable is set in db_impl_readonly.cc and it's actually DBImpl::mem_. Why don't we just apply updates to mem_ right away? It seems more intuitive. * Changes when we apply updates to manifest. Before, the process is to recover all the logs, flush it to sst files and then do one giant commit that atomically adds all recovered sst files and sets the next log number. This works good enough, but causes some small troubles for my column family approach, since I can't have one VersionEdit apply to more than single column family[1]. The change here is to commit the files recovered from logs right away. Here is the state of the world before the change: 1. Recover log 5, add new sst files to edit 2. Recover log 7, add new sst files to edit 3. Recover log 8, add new sst files to edit 4. Commit all added sst files to manifest and mark log files 5, 7 and 8 as recoverd (via SetLogNumber(9) function) After the change, we'll do: 1. Recover log 5, commit the new sst files and set log 5 as recovered 2. Recover log 7, commit the new sst files and set log 7 as recovered 3. Recover log 8, commit the new sst files and set log 8 as recovered The added (small) benefit is that if we fail after (2), the new recovery will only have to recover log 8. In previous case, we'll have to restart the recovery from the beginning. The bigger benefit will be to enable easier integration of multiple column families in Recovery code path. [1] I'm happy to dicuss this decison, but I believe this is the cleanest way to go. It also makes backward compatibility much easier. We don't have a requirement of adding multiple column families atomically. Test Plan: make check Reviewers: dhruba, haobo, kailiu, sdong Reviewed By: kailiu CC: leveldb Differential Revision: https://reviews.facebook.net/D15237
11 years ago
return status;
}
Status DBImpl::WriteLevel0TableForRecovery(ColumnFamilyData* cfd, MemTable* mem,
VersionEdit* edit) {
mutex_.AssertHeld();
const uint64_t start_micros = env_->NowMicros();
FileMetaData meta;
meta.fd = FileDescriptor(versions_->NewFileNumber(), 0, 0);
pending_outputs_[meta.fd.GetNumber()] = 0; // path 0 for level 0 file.
ReadOptions ro;
ro.total_order_seek = true;
Arena arena;
Status s;
{
ScopedArenaIterator iter(mem->NewIterator(ro, &arena));
const SequenceNumber newest_snapshot = snapshots_.GetNewest();
const SequenceNumber earliest_seqno_in_memtable =
mem->GetFirstSequenceNumber();
Log(db_options_.info_log, "[%s] Level-0 table #%" PRIu64 ": started",
cfd->GetName().c_str(), meta.fd.GetNumber());
{
mutex_.Unlock();
s = BuildTable(
dbname_, env_, *cfd->ioptions(), env_options_, cfd->table_cache(),
iter.get(), &meta, cfd->internal_comparator(), newest_snapshot,
earliest_seqno_in_memtable, GetCompressionFlush(*cfd->ioptions()),
cfd->ioptions()->compression_opts, Env::IO_HIGH);
LogFlush(db_options_.info_log);
mutex_.Lock();
}
Log(db_options_.info_log,
"[%s] Level-0 table #%" PRIu64 ": %" PRIu64 " bytes %s",
cfd->GetName().c_str(), meta.fd.GetNumber(), meta.fd.GetFileSize(),
s.ToString().c_str());
}
pending_outputs_.erase(meta.fd.GetNumber());
// Note that if file_size is zero, the file has been deleted and
// should not be added to the manifest.
int level = 0;
if (s.ok() && meta.fd.GetFileSize() > 0) {
edit->AddFile(level, meta.fd.GetNumber(), meta.fd.GetPathId(),
meta.fd.GetFileSize(), meta.smallest, meta.largest,
meta.smallest_seqno, meta.largest_seqno);
}
InternalStats::CompactionStats stats(1);
stats.micros = env_->NowMicros() - start_micros;
stats.bytes_written = meta.fd.GetFileSize();
Improve statistics Summary: This adds more statistics to be reported by GetProperty("leveldb.stats"). The new stats include time spent waiting on stalls in MakeRoomForWrite. This also includes the total amplification rate where that is: (#bytes of sequential IO during compaction) / (#bytes from Put) This also includes a lot more data for the per-level compaction report. * Rn(MB) - MB read from level N during compaction between levels N and N+1 * Rnp1(MB) - MB read from level N+1 during compaction between levels N and N+1 * Wnew(MB) - new data written to the level during compaction * Amplify - ( Write(MB) + Rnp1(MB) ) / Rn(MB) * Rn - files read from level N during compaction between levels N and N+1 * Rnp1 - files read from level N+1 during compaction between levels N and N+1 * Wnp1 - files written to level N+1 during compaction between levels N and N+1 * NewW - new files written to level N+1 during compaction * Count - number of compactions done for this level This is the new output from DB::GetProperty("leveldb.stats"). The old output stopped at Write(MB) Compactions Level Files Size(MB) Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count ------------------------------------------------------------------------------------------------------------------------------------- 0 3 6 33 0 576 0 0 576 -1.0 0.0 1.3 0 0 0 0 290 1 127 242 351 5316 5314 570 4747 567 17.0 12.1 12.1 287 2399 2685 286 32 2 161 328 54 822 824 326 496 328 4.0 1.9 1.9 160 251 411 160 161 Amplification: 22.3 rate, 0.56 GB in, 12.55 GB out Uptime(secs): 439.8 Stalls(secs): 206.938 level0_slowdown, 0.000 level0_numfiles, 24.129 memtable_compaction Task ID: # Blame Rev: Test Plan: run db_bench Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - (cherry picked from commit ecdeead38f86cc02e754d0032600742c4f02fec8) Reviewers: dhruba Differential Revision: https://reviews.facebook.net/D6153
12 years ago
stats.files_out_levelnp1 = 1;
cfd->internal_stats()->AddCompactionStats(level, stats);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
cfd->internal_stats()->AddCFStats(
InternalStats::BYTES_FLUSHED, meta.fd.GetFileSize());
RecordTick(stats_, COMPACT_WRITE_BYTES, meta.fd.GetFileSize());
return s;
}
Status DBImpl::FlushMemTableToOutputFile(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
bool* madeProgress, JobContext* job_context, LogBuffer* log_buffer) {
mutex_.AssertHeld();
assert(cfd->imm()->size() != 0);
assert(cfd->imm()->IsFlushPending());
FlushJob flush_job(dbname_, cfd, db_options_, mutable_cf_options,
env_options_, versions_.get(), &mutex_, &shutting_down_,
&pending_outputs_, snapshots_.GetNewest(), job_context,
log_buffer, db_directory_.get(),
GetCompressionFlush(*cfd->ioptions()), stats_);
Status s = flush_job.Run();
if (s.ok()) {
InstallSuperVersionBackground(cfd, job_context, mutable_cf_options);
if (madeProgress) {
*madeProgress = 1;
}
VersionStorageInfo::LevelSummaryStorage tmp;
LogToBuffer(log_buffer, "[%s] Level summary: %s\n", cfd->GetName().c_str(),
cfd->current()->GetStorageInfo()->LevelSummary(&tmp));
if (disable_delete_obsolete_files_ == 0) {
// add to deletion state
while (alive_log_files_.size() &&
alive_log_files_.begin()->number < versions_->MinLogNumber()) {
const auto& earliest = *alive_log_files_.begin();
job_context->log_delete_files.push_back(earliest.number);
total_log_size_ -= earliest.size;
alive_log_files_.pop_front();
}
}
}
if (!s.ok() && !s.IsShutdownInProgress() && db_options_.paranoid_checks &&
bg_error_.ok()) {
// if a bad error happened (not ShutdownInProgress) and paranoid_checks is
// true, mark DB read-only
bg_error_ = s;
}
RecordFlushIOStats();
return s;
}
Status DBImpl::CompactRange(ColumnFamilyHandle* column_family,
const Slice* begin, const Slice* end,
bool reduce_level, int target_level,
uint32_t target_path_id) {
if (target_path_id >= db_options_.db_paths.size()) {
return Status::InvalidArgument("Invalid target path ID");
}
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
Status s = FlushMemTable(cfd, FlushOptions());
if (!s.ok()) {
LogFlush(db_options_.info_log);
return s;
}
int max_level_with_files = 0;
{
MutexLock l(&mutex_);
Version* base = cfd->current();
for (int level = 1; level < cfd->NumberLevels(); level++) {
if (base->GetStorageInfo()->OverlapInLevel(level, begin, end)) {
max_level_with_files = level;
}
}
}
for (int level = 0; level <= max_level_with_files; level++) {
// in case the compaction is unversal or if we're compacting the
// bottom-most level, the output level will be the same as input one.
// level 0 can never be the bottommost level (i.e. if all files are in level
// 0, we will compact to level 1)
if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO ||
(level == max_level_with_files && level > 0)) {
s = RunManualCompaction(cfd, level, level, target_path_id, begin, end);
} else {
s = RunManualCompaction(cfd, level, level + 1, target_path_id, begin,
end);
}
if (!s.ok()) {
LogFlush(db_options_.info_log);
return s;
}
}
if (reduce_level) {
s = ReFitLevel(cfd, max_level_with_files, target_level);
}
LogFlush(db_options_.info_log);
{
MutexLock l(&mutex_);
// an automatic compaction that has been scheduled might have been
// preempted by the manual compactions. Need to schedule it back.
MaybeScheduleFlushOrCompaction();
}
return s;
}
bool DBImpl::SetOptions(ColumnFamilyHandle* column_family,
const std::unordered_map<std::string, std::string>& options_map) {
auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family)->cfd();
if (options_map.empty()) {
Log(db_options_.info_log, "SetOptions() on column family [%s], empty input",
cfd->GetName().c_str());
return false;
}
MutableCFOptions new_options;
bool succeed = false;
{
MutexLock l(&mutex_);
if (cfd->SetOptions(options_map)) {
new_options = *cfd->GetLatestMutableCFOptions();
succeed = true;
}
}
Log(db_options_.info_log, "SetOptions() on column family [%s], inputs:",
cfd->GetName().c_str());
for (const auto& o : options_map) {
Log(db_options_.info_log, "%s: %s\n", o.first.c_str(), o.second.c_str());
}
if (succeed) {
Log(db_options_.info_log, "[%s] SetOptions succeeded",
cfd->GetName().c_str());
new_options.Dump(db_options_.info_log.get());
} else {
Log(db_options_.info_log, "[%s] SetOptions failed",
cfd->GetName().c_str());
}
return succeed;
}
// return the same level if it cannot be moved
int DBImpl::FindMinimumEmptyLevelFitting(ColumnFamilyData* cfd,
const MutableCFOptions& mutable_cf_options, int level) {
mutex_.AssertHeld();
auto* vstorage = cfd->current()->GetStorageInfo();
int minimum_level = level;
for (int i = level - 1; i > 0; --i) {
// stop if level i is not empty
if (vstorage->NumLevelFiles(i) > 0) break;
// stop if level i is too small (cannot fit the level files)
if (mutable_cf_options.MaxBytesForLevel(i) <
vstorage->NumLevelBytes(level)) {
break;
}
minimum_level = i;
}
return minimum_level;
}
Status DBImpl::ReFitLevel(ColumnFamilyData* cfd, int level, int target_level) {
assert(level < cfd->NumberLevels());
SuperVersion* superversion_to_free = nullptr;
SuperVersion* new_superversion = new SuperVersion();
mutex_.Lock();
// only allow one thread refitting
if (refitting_level_) {
mutex_.Unlock();
Log(db_options_.info_log, "ReFitLevel: another thread is refitting");
delete new_superversion;
return Status::NotSupported("another thread is refitting");
}
refitting_level_ = true;
// wait for all background threads to stop
bg_work_gate_closed_ = true;
while (bg_compaction_scheduled_ > 0 || bg_flush_scheduled_) {
Log(db_options_.info_log,
"RefitLevel: waiting for background threads to stop: %d %d",
bg_compaction_scheduled_, bg_flush_scheduled_);
bg_cv_.Wait();
}
const MutableCFOptions mutable_cf_options =
*cfd->GetLatestMutableCFOptions();
// move to a smaller level
int to_level = target_level;
if (target_level < 0) {
to_level = FindMinimumEmptyLevelFitting(cfd, mutable_cf_options, level);
}
assert(to_level <= level);
Status status;
if (to_level < level) {
Log(db_options_.info_log, "[%s] Before refitting:\n%s",
cfd->GetName().c_str(), cfd->current()->DebugString().data());
VersionEdit edit;
edit.SetColumnFamily(cfd->GetID());
for (const auto& f : cfd->current()->GetStorageInfo()->files_[level]) {
edit.DeleteFile(level, f->fd.GetNumber());
edit.AddFile(to_level, f->fd.GetNumber(), f->fd.GetPathId(),
f->fd.GetFileSize(), f->smallest, f->largest,
f->smallest_seqno, f->largest_seqno);
}
Log(db_options_.info_log, "[%s] Apply version edit:\n%s",
cfd->GetName().c_str(), edit.DebugString().data());
status = versions_->LogAndApply(cfd,
mutable_cf_options, &edit, &mutex_, db_directory_.get());
superversion_to_free = InstallSuperVersion(
cfd, new_superversion, mutable_cf_options);
new_superversion = nullptr;
Log(db_options_.info_log, "[%s] LogAndApply: %s\n", cfd->GetName().c_str(),
status.ToString().data());
if (status.ok()) {
Log(db_options_.info_log, "[%s] After refitting:\n%s",
cfd->GetName().c_str(), cfd->current()->DebugString().data());
}
}
refitting_level_ = false;
bg_work_gate_closed_ = false;
mutex_.Unlock();
delete superversion_to_free;
delete new_superversion;
return status;
}
int DBImpl::NumberLevels(ColumnFamilyHandle* column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
return cfh->cfd()->NumberLevels();
}
int DBImpl::MaxMemCompactionLevel(ColumnFamilyHandle* column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
MutexLock l(&mutex_);
return cfh->cfd()->GetSuperVersion()->
mutable_cf_options.max_mem_compaction_level;
}
int DBImpl::Level0StopWriteTrigger(ColumnFamilyHandle* column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
MutexLock l(&mutex_);
return cfh->cfd()->GetSuperVersion()->
mutable_cf_options.level0_stop_writes_trigger;
}
Status DBImpl::Flush(const FlushOptions& flush_options,
ColumnFamilyHandle* column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
return FlushMemTable(cfh->cfd(), flush_options);
}
SequenceNumber DBImpl::GetLatestSequenceNumber() const {
return versions_->LastSequence();
}
Status DBImpl::RunManualCompaction(ColumnFamilyData* cfd, int input_level,
int output_level, uint32_t output_path_id,
const Slice* begin, const Slice* end) {
assert(input_level >= 0);
InternalKey begin_storage, end_storage;
ManualCompaction manual;
manual.cfd = cfd;
manual.input_level = input_level;
manual.output_level = output_level;
manual.output_path_id = output_path_id;
manual.done = false;
manual.in_progress = false;
// For universal compaction, we enforce every manual compaction to compact
// all files.
if (begin == nullptr ||
cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
manual.begin = nullptr;
} else {
begin_storage = InternalKey(*begin, kMaxSequenceNumber, kValueTypeForSeek);
manual.begin = &begin_storage;
}
if (end == nullptr ||
cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
manual.end = nullptr;
} else {
end_storage = InternalKey(*end, 0, static_cast<ValueType>(0));
manual.end = &end_storage;
}
MutexLock l(&mutex_);
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
// When a manual compaction arrives, temporarily disable scheduling of
// non-manual compactions and wait until the number of scheduled compaction
// jobs drops to zero. This is needed to ensure that this manual compaction
// can compact any range of keys/files.
//
// bg_manual_only_ is non-zero when at least one thread is inside
// RunManualCompaction(), i.e. during that time no other compaction will
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
// get scheduled (see MaybeScheduleFlushOrCompaction).
//
// Note that the following loop doesn't stop more that one thread calling
// RunManualCompaction() from getting to the second while loop below.
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
// However, only one of them will actually schedule compaction, while
// others will wait on a condition variable until it completes.
++bg_manual_only_;
while (bg_compaction_scheduled_ > 0) {
Log(db_options_.info_log,
"[%s] Manual compaction waiting for all other scheduled background "
"compactions to finish",
cfd->GetName().c_str());
bg_cv_.Wait();
}
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
Log(db_options_.info_log, "[%s] Manual compaction starting",
cfd->GetName().c_str());
// We don't check bg_error_ here, because if we get the error in compaction,
// the compaction will set manual.status to bg_error_ and set manual.done to
// true.
while (!manual.done) {
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
assert(bg_manual_only_ > 0);
if (manual_compaction_ != nullptr) {
// Running either this or some other manual compaction
bg_cv_.Wait();
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
} else {
manual_compaction_ = &manual;
bg_compaction_scheduled_++;
env_->Schedule(&DBImpl::BGWorkCompaction, this, Env::Priority::LOW);
}
}
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
11 years ago
assert(!manual.in_progress);
assert(bg_manual_only_ > 0);
--bg_manual_only_;
return manual.status;
}
Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
const FlushOptions& flush_options) {
Status s;
{
WriteContext context;
MutexLock guard_lock(&mutex_);
if (cfd->imm()->size() == 0 && cfd->mem()->IsEmpty()) {
// Nothing to flush
return Status::OK();
}
WriteThread::Writer w(&mutex_);
s = write_thread_.EnterWriteThread(&w, 0);
assert(s.ok() && !w.done); // No timeout and nobody should do our job
// SetNewMemtableAndNewLogFile() will release and reacquire mutex
// during execution
s = SetNewMemtableAndNewLogFile(cfd, &context);
cfd->imm()->FlushRequested();
MaybeScheduleFlushOrCompaction();
write_thread_.ExitWriteThread(&w, &w, s);
}
if (s.ok() && flush_options.wait) {
// Wait until the compaction completes
s = WaitForFlushMemTable(cfd);
}
return s;
}
Status DBImpl::WaitForFlushMemTable(ColumnFamilyData* cfd) {
Status s;
// Wait until the compaction completes
MutexLock l(&mutex_);
while (cfd->imm()->size() > 0 && bg_error_.ok()) {
bg_cv_.Wait();
}
if (!bg_error_.ok()) {
s = bg_error_;
}
return s;
}
void DBImpl::MaybeScheduleFlushOrCompaction() {
mutex_.AssertHeld();
bg_schedule_needed_ = false;
if (bg_work_gate_closed_) {
// gate closed for backgrond work
} else if (shutting_down_.load(std::memory_order_acquire)) {
// DB is being deleted; no more background compactions
} else {
bool is_flush_pending = false;
// no need to refcount since we're under a mutex
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->imm()->IsFlushPending()) {
is_flush_pending = true;
}
}
if (is_flush_pending) {
// memtable flush needed
if (bg_flush_scheduled_ < db_options_.max_background_flushes) {
bg_flush_scheduled_++;
env_->Schedule(&DBImpl::BGWorkFlush, this, Env::Priority::HIGH);
} else if (db_options_.max_background_flushes > 0) {
bg_schedule_needed_ = true;
}
}
bool is_compaction_needed = false;
// no need to refcount since we're under a mutex
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->current()->GetStorageInfo()->NeedsCompaction()) {
is_compaction_needed = true;
break;
}
}
// Schedule BGWorkCompaction if there's a compaction pending (or a memtable
// flush, but the HIGH pool is not enabled)
// Do it only if max_background_compactions hasn't been reached and
// bg_manual_only_ == 0
if (!bg_manual_only_ &&
(is_compaction_needed ||
(is_flush_pending && db_options_.max_background_flushes == 0))) {
if (bg_compaction_scheduled_ < db_options_.max_background_compactions) {
bg_compaction_scheduled_++;
env_->Schedule(&DBImpl::BGWorkCompaction, this, Env::Priority::LOW);
} else {
bg_schedule_needed_ = true;
}
}
}
}
void DBImpl::RecordFlushIOStats() {
RecordTick(stats_, FLUSH_WRITE_BYTES, IOSTATS(bytes_written));
IOSTATS_RESET(bytes_written);
}
void DBImpl::RecordCompactionIOStats() {
RecordTick(stats_, COMPACT_READ_BYTES, IOSTATS(bytes_read));
IOSTATS_RESET(bytes_read);
RecordTick(stats_, COMPACT_WRITE_BYTES, IOSTATS(bytes_written));
IOSTATS_RESET(bytes_written);
}
void DBImpl::BGWorkFlush(void* db) {
IOSTATS_SET_THREAD_POOL_ID(Env::Priority::HIGH);
reinterpret_cast<DBImpl*>(db)->BackgroundCallFlush();
}
void DBImpl::BGWorkCompaction(void* db) {
IOSTATS_SET_THREAD_POOL_ID(Env::Priority::LOW);
reinterpret_cast<DBImpl*>(db)->BackgroundCallCompaction();
}
Status DBImpl::BackgroundFlush(bool* madeProgress, JobContext* job_context,
LogBuffer* log_buffer) {
mutex_.AssertHeld();
if (!bg_error_.ok()) {
return bg_error_;
}
// call_status is failure if at least one flush was a failure. even if
// flushing one column family reports a failure, we will continue flushing
// other column families. however, call_status will be a failure in that case.
Status call_status;
// refcounting in iteration
for (auto cfd : *versions_->GetColumnFamilySet()) {
cfd->Ref();
Status flush_status;
const MutableCFOptions mutable_cf_options =
*cfd->GetLatestMutableCFOptions();
while (flush_status.ok() && cfd->imm()->IsFlushPending()) {
LogToBuffer(
log_buffer,
"BackgroundCallFlush doing FlushMemTableToOutputFile with column "
"family [%s], flush slots available %d",
cfd->GetName().c_str(),
db_options_.max_background_flushes - bg_flush_scheduled_);
flush_status = FlushMemTableToOutputFile(
cfd, mutable_cf_options, madeProgress, job_context, log_buffer);
}
if (call_status.ok() && !flush_status.ok()) {
call_status = flush_status;
}
cfd->Unref();
}
versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
return call_status;
}
void DBImpl::BackgroundCallFlush() {
bool madeProgress = false;
JobContext job_context(true);
assert(bg_flush_scheduled_);
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
{
MutexLock l(&mutex_);
Status s;
if (!shutting_down_.load(std::memory_order_acquire)) {
s = BackgroundFlush(&madeProgress, &job_context, &log_buffer);
if (!s.ok()) {
// Wait a little bit before retrying background compaction in
// case this is an environmental problem and we do not want to
// chew up resources for failed compactions for the duration of
// the problem.
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
uint64_t error_cnt =
default_cf_internal_stats_->BumpAndGetBackgroundErrorCount();
bg_cv_.SignalAll(); // In case a waiter can proceed despite the error
mutex_.Unlock();
Log(db_options_.info_log,
"Waiting after background flush error: %s"
"Accumulated background error counts: %" PRIu64,
s.ToString().c_str(), error_cnt);
log_buffer.FlushBufferToLog();
LogFlush(db_options_.info_log);
env_->SleepForMicroseconds(1000000);
mutex_.Lock();
}
}
// If !s.ok(), this means that Flush failed. In that case, we want
// to delete all obsolete files and we force FindObsoleteFiles()
FindObsoleteFiles(&job_context, !s.ok());
// delete unnecessary files if any, this is done outside the mutex
if (job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
mutex_.Unlock();
// Have to flush the info logs before bg_flush_scheduled_--
// because if bg_flush_scheduled_ becomes 0 and the lock is
// released, the deconstructor of DB can kick in and destroy all the
// states of DB so info_log might not be available after that point.
// It also applies to access other states that DB owns.
log_buffer.FlushBufferToLog();
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
mutex_.Lock();
}
bg_flush_scheduled_--;
// Any time the mutex is released After finding the work to do, another
// thread might execute MaybeScheduleFlushOrCompaction(). It is possible
// that there is a pending job but it is not scheduled because of the
// max thread limit.
if (madeProgress || bg_schedule_needed_) {
MaybeScheduleFlushOrCompaction();
}
RecordFlushIOStats();
bg_cv_.SignalAll();
// IMPORTANT: there should be no code after calling SignalAll. This call may
// signal the DB destructor that it's OK to proceed with destruction. In
// that case, all DB variables will be dealloacated and referencing them
// will cause trouble.
}
}
void DBImpl::BackgroundCallCompaction() {
bool madeProgress = false;
JobContext job_context(true);
MaybeDumpStats();
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
{
MutexLock l(&mutex_);
assert(bg_compaction_scheduled_);
Status s;
if (!shutting_down_.load(std::memory_order_acquire)) {
s = BackgroundCompaction(&madeProgress, &job_context, &log_buffer);
if (!s.ok()) {
// Wait a little bit before retrying background compaction in
// case this is an environmental problem and we do not want to
// chew up resources for failed compactions for the duration of
// the problem.
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
uint64_t error_cnt =
default_cf_internal_stats_->BumpAndGetBackgroundErrorCount();
bg_cv_.SignalAll(); // In case a waiter can proceed despite the error
mutex_.Unlock();
log_buffer.FlushBufferToLog();
Log(db_options_.info_log,
"Waiting after background compaction error: %s, "
"Accumulated background error counts: %" PRIu64,
s.ToString().c_str(), error_cnt);
LogFlush(db_options_.info_log);
env_->SleepForMicroseconds(1000000);
mutex_.Lock();
}
}
// If !s.ok(), this means that Compaction failed. In that case, we want
// to delete all obsolete files we might have created and we force
// FindObsoleteFiles(). This is because job_context does not
// catch all created files if compaction failed.
FindObsoleteFiles(&job_context, !s.ok());
// delete unnecessary files if any, this is done outside the mutex
if (job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
mutex_.Unlock();
// Have to flush the info logs before bg_compaction_scheduled_--
// because if bg_flush_scheduled_ becomes 0 and the lock is
// released, the deconstructor of DB can kick in and destroy all the
// states of DB so info_log might not be available after that point.
// It also applies to access other states that DB owns.
log_buffer.FlushBufferToLog();
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
mutex_.Lock();
}
bg_compaction_scheduled_--;
versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
// Previous compaction may have produced too many files in a level,
// So reschedule another compaction if we made progress in the
// last compaction.
//
// Also, any time the mutex is released After finding the work to do,
// another thread might execute MaybeScheduleFlushOrCompaction(). It is
// possible that there is a pending job but it is not scheduled because of
// the max thread limit.
if (madeProgress || bg_schedule_needed_) {
MaybeScheduleFlushOrCompaction();
}
if (madeProgress || bg_compaction_scheduled_ == 0 || bg_manual_only_ > 0) {
// signal if
// * madeProgress -- need to wakeup DelayWrite
// * bg_compaction_scheduled_ == 0 -- need to wakeup ~DBImpl
// * bg_manual_only_ > 0 -- need to wakeup RunManualCompaction
// If none of this is true, there is no need to signal since nobody is
// waiting for it
bg_cv_.SignalAll();
}
// IMPORTANT: there should be no code after calling SignalAll. This call may
// signal the DB destructor that it's OK to proceed with destruction. In
// that case, all DB variables will be dealloacated and referencing them
// will cause trouble.
}
}
Status DBImpl::BackgroundCompaction(bool* madeProgress, JobContext* job_context,
LogBuffer* log_buffer) {
*madeProgress = false;
mutex_.AssertHeld();
bool is_manual = (manual_compaction_ != nullptr) &&
(manual_compaction_->in_progress == false);
if (!bg_error_.ok()) {
if (is_manual) {
manual_compaction_->status = bg_error_;
manual_compaction_->done = true;
manual_compaction_->in_progress = false;
manual_compaction_ = nullptr;
}
return bg_error_;
}
if (is_manual) {
// another thread cannot pick up the same work
manual_compaction_->in_progress = true;
} else if (manual_compaction_ != nullptr) {
// there should be no automatic compactions running when manual compaction
// is running
return Status::OK();
}
// FLUSH preempts compaction
Status flush_stat;
for (auto cfd : *versions_->GetColumnFamilySet()) {
const MutableCFOptions mutable_cf_options =
*cfd->GetLatestMutableCFOptions();
while (cfd->imm()->IsFlushPending()) {
LogToBuffer(
log_buffer,
"BackgroundCompaction doing FlushMemTableToOutputFile, "
"compaction slots available %d",
db_options_.max_background_compactions - bg_compaction_scheduled_);
cfd->Ref();
flush_stat = FlushMemTableToOutputFile(
cfd, mutable_cf_options, madeProgress, job_context, log_buffer);
cfd->Unref();
if (!flush_stat.ok()) {
if (is_manual) {
manual_compaction_->status = flush_stat;
manual_compaction_->done = true;
manual_compaction_->in_progress = false;
manual_compaction_ = nullptr;
}
return flush_stat;
}
}
}
// Compaction makes a copy of the latest MutableCFOptions. It should be used
// throughout the compaction procedure to make sure consistency. It will
// eventually be installed into SuperVersion
unique_ptr<Compaction> c;
InternalKey manual_end_storage;
InternalKey* manual_end = &manual_end_storage;
if (is_manual) {
ManualCompaction* m = manual_compaction_;
assert(m->in_progress);
c.reset(m->cfd->CompactRange(
*m->cfd->GetLatestMutableCFOptions(), m->input_level, m->output_level,
m->output_path_id, m->begin, m->end, &manual_end));
if (!c) {
m->done = true;
}
LogToBuffer(log_buffer,
"[%s] Manual compaction from level-%d to level-%d from %s .. "
"%s; will stop at %s\n",
m->cfd->GetName().c_str(), m->input_level, m->output_level,
(m->begin ? m->begin->DebugString().c_str() : "(begin)"),
(m->end ? m->end->DebugString().c_str() : "(end)"),
((m->done || manual_end == nullptr)
? "(end)"
: manual_end->DebugString().c_str()));
} else {
// no need to refcount in iteration since it's always under a mutex
for (auto cfd : *versions_->GetColumnFamilySet()) {
// Pick up latest mutable CF Options and use it throughout the
// compaction job
auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
if (!mutable_cf_options->disable_auto_compactions) {
// NOTE: try to avoid unnecessary copy of MutableCFOptions if
// compaction is not necessary. Need to make sure mutex is held
// until we make a copy in the following code
c.reset(cfd->PickCompaction(*mutable_cf_options, log_buffer));
if (c != nullptr) {
// update statistics
MeasureTime(stats_, NUM_FILES_IN_SINGLE_COMPACTION,
c->inputs(0)->size());
break;
}
}
}
}
Status status;
if (!c) {
// Nothing to do
LogToBuffer(log_buffer, "Compaction nothing to do");
} else if (c->IsDeletionCompaction()) {
// TODO(icanadi) Do we want to honor snapshots here? i.e. not delete old
// file if there is alive snapshot pointing to it
assert(c->num_input_files(1) == 0);
assert(c->level() == 0);
assert(c->column_family_data()->ioptions()->compaction_style ==
kCompactionStyleFIFO);
for (const auto& f : *c->inputs(0)) {
c->edit()->DeleteFile(c->level(), f->fd.GetNumber());
}
status = versions_->LogAndApply(
c->column_family_data(), *c->mutable_cf_options(), c->edit(),
&mutex_, db_directory_.get());
InstallSuperVersionBackground(c->column_family_data(), job_context,
*c->mutable_cf_options());
LogToBuffer(log_buffer, "[%s] Deleted %d files\n",
c->column_family_data()->GetName().c_str(),
c->num_input_files(0));
c->ReleaseCompactionFiles(status);
*madeProgress = true;
} else if (!is_manual && c->IsTrivialMove()) {
// Move file to next level
assert(c->num_input_files(0) == 1);
FileMetaData* f = c->input(0, 0);
c->edit()->DeleteFile(c->level(), f->fd.GetNumber());
c->edit()->AddFile(c->level() + 1, f->fd.GetNumber(), f->fd.GetPathId(),
f->fd.GetFileSize(), f->smallest, f->largest,
f->smallest_seqno, f->largest_seqno);
status = versions_->LogAndApply(c->column_family_data(),
*c->mutable_cf_options(),
c->edit(), &mutex_, db_directory_.get());
// Use latest MutableCFOptions
InstallSuperVersionBackground(c->column_family_data(), job_context,
*c->mutable_cf_options());
VersionStorageInfo::LevelSummaryStorage tmp;
LogToBuffer(log_buffer, "[%s] Moved #%" PRIu64 " to level-%d %" PRIu64
" bytes %s: %s\n",
c->column_family_data()->GetName().c_str(), f->fd.GetNumber(),
c->level() + 1, f->fd.GetFileSize(), status.ToString().c_str(),
c->input_version()->GetStorageInfo()->LevelSummary(&tmp));
c->ReleaseCompactionFiles(status);
*madeProgress = true;
} else {
MaybeScheduleFlushOrCompaction(); // do more compaction work in parallel.
CompactionState* compact = new CompactionState(c.get());
status = DoCompactionWork(compact, *c->mutable_cf_options(), job_context,
log_buffer);
CleanupCompaction(compact, status);
c->ReleaseCompactionFiles(status);
c->ReleaseInputs();
*madeProgress = true;
}
c.reset();
if (status.ok()) {
// Done
} else if (status.IsShutdownInProgress()) {
// Ignore compaction errors found during shutting down
} else {
Log(InfoLogLevel::WARN_LEVEL, db_options_.info_log, "Compaction error: %s",
status.ToString().c_str());
if (db_options_.paranoid_checks && bg_error_.ok()) {
bg_error_ = status;
}
}
if (is_manual) {
ManualCompaction* m = manual_compaction_;
if (!status.ok()) {
m->status = status;
m->done = true;
}
// For universal compaction:
// Because universal compaction always happens at level 0, so one
// compaction will pick up all overlapped files. No files will be
// filtered out due to size limit and left for a successive compaction.
// So we can safely conclude the current compaction.
//
// Also note that, if we don't stop here, then the current compaction
// writes a new file back to level 0, which will be used in successive
// compaction. Hence the manual compaction will never finish.
//
// Stop the compaction if manual_end points to nullptr -- this means
// that we compacted the whole range. manual_end should always point
// to nullptr in case of universal compaction
if (manual_end == nullptr) {
m->done = true;
}
if (!m->done) {
// We only compacted part of the requested range. Update *m
// to the range that is left to be compacted.
// Universal and FIFO compactions should always compact the whole range
assert(m->cfd->ioptions()->compaction_style != kCompactionStyleUniversal);
assert(m->cfd->ioptions()->compaction_style != kCompactionStyleFIFO);
m->tmp_storage = *manual_end;
m->begin = &m->tmp_storage;
}
m->in_progress = false; // not being processed anymore
manual_compaction_ = nullptr;
}
return status;
}
void DBImpl::CleanupCompaction(CompactionState* compact, Status status) {
mutex_.AssertHeld();
if (compact->builder != nullptr) {
// May happen if we get a shutdown call in the middle of compaction
compact->builder->Abandon();
compact->builder.reset();
} else {
assert(!status.ok() || compact->outfile == nullptr);
}
for (size_t i = 0; i < compact->outputs.size(); i++) {
const CompactionState::Output& out = compact->outputs[i];
pending_outputs_.erase(out.number);
// If this file was inserted into the table cache then remove
// them here because this compaction was not committed.
if (!status.ok()) {
TableCache::Evict(table_cache_.get(), out.number);
}
}
delete compact;
}
// Allocate the file numbers for the output file. We allocate as
// many output file numbers as there are files in level+1 (at least one)
// Insert them into pending_outputs so that they do not get deleted.
void DBImpl::AllocateCompactionOutputFileNumbers(CompactionState* compact) {
mutex_.AssertHeld();
assert(compact != nullptr);
assert(compact->builder == nullptr);
int filesNeeded = compact->compaction->num_input_files(1);
for (int i = 0; i < std::max(filesNeeded, 1); i++) {
uint64_t file_number = versions_->NewFileNumber();
pending_outputs_[file_number] = compact->compaction->GetOutputPathId();
compact->allocated_file_numbers.push_back(file_number);
}
}
// Frees up unused file number.
void DBImpl::ReleaseCompactionUnusedFileNumbers(CompactionState* compact) {
mutex_.AssertHeld();
for (const auto file_number : compact->allocated_file_numbers) {
pending_outputs_.erase(file_number);
}
}
Status DBImpl::OpenCompactionOutputFile(
CompactionState* compact, const MutableCFOptions& mutable_cf_options) {
assert(compact != nullptr);
assert(compact->builder == nullptr);
uint64_t file_number;
// If we have not yet exhausted the pre-allocated file numbers,
// then use the one from the front. Otherwise, we have to acquire
// the heavyweight lock and allocate a new file number.
if (!compact->allocated_file_numbers.empty()) {
file_number = compact->allocated_file_numbers.front();
compact->allocated_file_numbers.pop_front();
} else {
mutex_.Lock();
file_number = versions_->NewFileNumber();
pending_outputs_[file_number] = compact->compaction->GetOutputPathId();
mutex_.Unlock();
}
// Make the output file
std::string fname = TableFileName(db_options_.db_paths, file_number,
compact->compaction->GetOutputPathId());
Status s = env_->NewWritableFile(fname, &compact->outfile, env_options_);
if (!s.ok()) {
Log(InfoLogLevel::ERROR_LEVEL, db_options_.info_log,
"[%s] OpenCompactionOutputFiles for table #%" PRIu64 " "
"fails at NewWritableFile with status %s",
compact->compaction->column_family_data()->GetName().c_str(),
file_number, s.ToString().c_str());
LogFlush(db_options_.info_log);
return s;
}
CompactionState::Output out;
out.number = file_number;
out.path_id = compact->compaction->GetOutputPathId();
out.smallest.Clear();
out.largest.Clear();
out.smallest_seqno = out.largest_seqno = 0;
compact->outputs.push_back(out);
compact->outfile->SetIOPriority(Env::IO_LOW);
compact->outfile->SetPreallocationBlockSize(
compact->compaction->OutputFilePreallocationSize(mutable_cf_options));
ColumnFamilyData* cfd = compact->compaction->column_family_data();
compact->builder.reset(NewTableBuilder(
*cfd->ioptions(), cfd->internal_comparator(), compact->outfile.get(),
compact->compaction->OutputCompressionType(),
cfd->ioptions()->compression_opts));
LogFlush(db_options_.info_log);
return s;
}
Status DBImpl::FinishCompactionOutputFile(CompactionState* compact,
Iterator* input) {
assert(compact != nullptr);
assert(compact->outfile);
assert(compact->builder != nullptr);
const uint64_t output_number = compact->current_output()->number;
const uint32_t output_path_id = compact->current_output()->path_id;
assert(output_number != 0);
// Check for iterator errors
Status s = input->status();
const uint64_t current_entries = compact->builder->NumEntries();
if (s.ok()) {
s = compact->builder->Finish();
} else {
compact->builder->Abandon();
}
const uint64_t current_bytes = compact->builder->FileSize();
compact->current_output()->file_size = current_bytes;
compact->total_bytes += current_bytes;
compact->builder.reset();
// Finish and check for file errors
if (s.ok() && !db_options_.disableDataSync) {
if (db_options_.use_fsync) {
StopWatch sw(env_, stats_, COMPACTION_OUTFILE_SYNC_MICROS);
s = compact->outfile->Fsync();
} else {
StopWatch sw(env_, stats_, COMPACTION_OUTFILE_SYNC_MICROS);
s = compact->outfile->Sync();
}
}
if (s.ok()) {
s = compact->outfile->Close();
}
compact->outfile.reset();
if (s.ok() && current_entries > 0) {
// Verify that the table is usable
ColumnFamilyData* cfd = compact->compaction->column_family_data();
FileDescriptor fd(output_number, output_path_id, current_bytes);
Iterator* iter = cfd->table_cache()->NewIterator(
ReadOptions(), env_options_, cfd->internal_comparator(), fd);
s = iter->status();
delete iter;
if (s.ok()) {
Log(db_options_.info_log, "[%s] Generated table #%" PRIu64 ": %" PRIu64
" keys, %" PRIu64 " bytes",
cfd->GetName().c_str(), output_number, current_entries,
current_bytes);
}
}
return s;
}
Status DBImpl::InstallCompactionResults(CompactionState* compact,
const MutableCFOptions& mutable_cf_options, LogBuffer* log_buffer) {
mutex_.AssertHeld();
// paranoia: verify that the files that we started with
// still exist in the current version and in the same original level.
// This ensures that a concurrent compaction did not erroneously
// pick the same files to compact.
if (!versions_->VerifyCompactionFileConsistency(compact->compaction)) {
Log(db_options_.info_log, "[%s] Compaction %d@%d + %d@%d files aborted",
compact->compaction->column_family_data()->GetName().c_str(),
compact->compaction->num_input_files(0), compact->compaction->level(),
compact->compaction->num_input_files(1),
compact->compaction->output_level());
return Status::Corruption("Compaction input files inconsistent");
}
LogToBuffer(log_buffer, "[%s] Compacted %d@%d + %d@%d files => %lld bytes",
compact->compaction->column_family_data()->GetName().c_str(),
compact->compaction->num_input_files(0),
compact->compaction->level(),
compact->compaction->num_input_files(1),
compact->compaction->output_level(),
static_cast<long long>(compact->total_bytes));
// Add compaction outputs
compact->compaction->AddInputDeletions(compact->compaction->edit());
for (size_t i = 0; i < compact->outputs.size(); i++) {
const CompactionState::Output& out = compact->outputs[i];
compact->compaction->edit()->AddFile(compact->compaction->output_level(),
out.number, out.path_id, out.file_size,
out.smallest, out.largest,
out.smallest_seqno, out.largest_seqno);
}
return versions_->LogAndApply(compact->compaction->column_family_data(),
mutable_cf_options,
compact->compaction->edit(), &mutex_,
db_directory_.get());
}
// Given a sequence number, return the sequence number of the
// earliest snapshot that this sequence number is visible in.
// The snapshots themselves are arranged in ascending order of
// sequence numbers.
// Employ a sequential search because the total number of
// snapshots are typically small.
inline SequenceNumber DBImpl::findEarliestVisibleSnapshot(
SequenceNumber in, std::vector<SequenceNumber>& snapshots,
SequenceNumber* prev_snapshot) {
SequenceNumber prev __attribute__((unused)) = 0;
for (const auto cur : snapshots) {
assert(prev <= cur);
if (cur >= in) {
*prev_snapshot = prev;
return cur;
}
prev = cur; // assignment
assert(prev);
}
Log(db_options_.info_log,
"Looking for seqid %" PRIu64 " but maxseqid is %" PRIu64 "", in,
snapshots[snapshots.size() - 1]);
assert(0);
return 0;
}
uint64_t DBImpl::CallFlushDuringCompaction(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
JobContext* job_context, LogBuffer* log_buffer) {
if (db_options_.max_background_flushes > 0) {
// flush thread will take care of this
return 0;
}
if (cfd->imm()->imm_flush_needed.load(std::memory_order_relaxed)) {
const uint64_t imm_start = env_->NowMicros();
mutex_.Lock();
if (cfd->imm()->IsFlushPending()) {
cfd->Ref();
FlushMemTableToOutputFile(cfd, mutable_cf_options, nullptr, job_context,
log_buffer);
cfd->Unref();
bg_cv_.SignalAll(); // Wakeup DelayWrite() if necessary
}
mutex_.Unlock();
log_buffer->FlushBufferToLog();
return env_->NowMicros() - imm_start;
}
return 0;
}
Status DBImpl::ProcessKeyValueCompaction(
const MutableCFOptions& mutable_cf_options, bool is_snapshot_supported,
SequenceNumber visible_at_tip, SequenceNumber earliest_snapshot,
SequenceNumber latest_snapshot, JobContext* job_context,
bool bottommost_level, int64_t* imm_micros, Iterator* input,
CompactionState* compact, bool is_compaction_v2, int* num_output_records,
LogBuffer* log_buffer) {
assert(num_output_records != nullptr);
size_t combined_idx = 0;
Status status;
std::string compaction_filter_value;
ParsedInternalKey ikey;
IterKey current_user_key;
bool has_current_user_key = false;
IterKey delete_key;
SequenceNumber last_sequence_for_key __attribute__((unused)) =
kMaxSequenceNumber;
SequenceNumber visible_in_snapshot = kMaxSequenceNumber;
ColumnFamilyData* cfd = compact->compaction->column_family_data();
MergeHelper merge(
cfd->user_comparator(), cfd->ioptions()->merge_operator,
db_options_.info_log.get(), cfd->ioptions()->min_partial_merge_operands,
false /* internal key corruption is expected */);
auto compaction_filter = cfd->ioptions()->compaction_filter;
std::unique_ptr<CompactionFilter> compaction_filter_from_factory = nullptr;
if (!compaction_filter) {
auto context = compact->GetFilterContextV1();
compaction_filter_from_factory =
cfd->ioptions()->compaction_filter_factory->CreateCompactionFilter(
context);
compaction_filter = compaction_filter_from_factory.get();
}
int64_t key_drop_user = 0;
int64_t key_drop_newer_entry = 0;
int64_t key_drop_obsolete = 0;
int64_t loop_cnt = 0;
while (input->Valid() && !shutting_down_.load(std::memory_order_acquire) &&
!cfd->IsDropped() && status.ok()) {
if (++loop_cnt > 1000) {
if (key_drop_user > 0) {
RecordTick(stats_, COMPACTION_KEY_DROP_USER, key_drop_user);
key_drop_user = 0;
}
if (key_drop_newer_entry > 0) {
RecordTick(stats_, COMPACTION_KEY_DROP_NEWER_ENTRY,
key_drop_newer_entry);
key_drop_newer_entry = 0;
}
if (key_drop_obsolete > 0) {
RecordTick(stats_, COMPACTION_KEY_DROP_OBSOLETE, key_drop_obsolete);
key_drop_obsolete = 0;
}
RecordCompactionIOStats();
loop_cnt = 0;
}
// FLUSH preempts compaction
// TODO(icanadi) this currently only checks if flush is necessary on
// compacting column family. we should also check if flush is necessary on
// other column families, too
(*imm_micros) += CallFlushDuringCompaction(cfd, mutable_cf_options,
job_context, log_buffer);
Slice key;
Slice value;
// If is_compaction_v2 is on, kv-pairs are reset to the prefix batch.
// This prefix batch should contain results after calling
// compaction_filter_v2.
//
// If is_compaction_v2 is off, this function will go through all the
// kv-pairs in input.
if (!is_compaction_v2) {
key = input->key();
value = input->value();
} else {
if (combined_idx >= compact->combined_key_buf_.size()) {
break;
}
assert(combined_idx < compact->combined_key_buf_.size());
key = compact->combined_key_buf_[combined_idx];
value = compact->combined_value_buf_[combined_idx];
++combined_idx;
}
if (compact->compaction->ShouldStopBefore(key) &&
compact->builder != nullptr) {
status = FinishCompactionOutputFile(compact, input);
if (!status.ok()) {
break;
}
}
// Handle key/value, add to state, etc.
bool drop = false;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
bool current_entry_is_merging = false;
if (!ParseInternalKey(key, &ikey)) {
// Do not hide error keys
// TODO: error key stays in db forever? Figure out the intention/rationale
// v10 error v8 : we cannot hide v8 even though it's pretty obvious.
current_user_key.Clear();
has_current_user_key = false;
last_sequence_for_key = kMaxSequenceNumber;
visible_in_snapshot = kMaxSequenceNumber;
} else {
if (!has_current_user_key ||
cfd->user_comparator()->Compare(ikey.user_key,
current_user_key.GetKey()) != 0) {
// First occurrence of this user key
current_user_key.SetKey(ikey.user_key);
has_current_user_key = true;
last_sequence_for_key = kMaxSequenceNumber;
visible_in_snapshot = kMaxSequenceNumber;
// apply the compaction filter to the first occurrence of the user key
if (compaction_filter && !is_compaction_v2 &&
ikey.type == kTypeValue &&
(visible_at_tip || ikey.sequence > latest_snapshot)) {
// If the user has specified a compaction filter and the sequence
// number is greater than any external snapshot, then invoke the
// filter.
// If the return value of the compaction filter is true, replace
// the entry with a delete marker.
bool value_changed = false;
compaction_filter_value.clear();
bool to_delete = compaction_filter->Filter(
compact->compaction->level(), ikey.user_key, value,
&compaction_filter_value, &value_changed);
if (to_delete) {
// make a copy of the original key and convert it to a delete
delete_key.SetInternalKey(ExtractUserKey(key), ikey.sequence,
kTypeDeletion);
// anchor the key again
key = delete_key.GetKey();
// needed because ikey is backed by key
ParseInternalKey(key, &ikey);
// no value associated with delete
value.clear();
++key_drop_user;
} else if (value_changed) {
value = compaction_filter_value;
}
}
}
// If there are no snapshots, then this kv affect visibility at tip.
// Otherwise, search though all existing snapshots to find
// the earlist snapshot that is affected by this kv.
SequenceNumber prev_snapshot = 0; // 0 means no previous snapshot
SequenceNumber visible = visible_at_tip ? visible_at_tip :
is_snapshot_supported ? findEarliestVisibleSnapshot(ikey.sequence,
compact->existing_snapshots, &prev_snapshot)
: 0;
if (visible_in_snapshot == visible) {
// If the earliest snapshot is which this key is visible in
// is the same as the visibily of a previous instance of the
// same key, then this kv is not visible in any snapshot.
// Hidden by an newer entry for same user key
// TODO: why not > ?
assert(last_sequence_for_key >= ikey.sequence);
drop = true; // (A)
++key_drop_newer_entry;
} else if (ikey.type == kTypeDeletion &&
ikey.sequence <= earliest_snapshot &&
compact->compaction->KeyNotExistsBeyondOutputLevel(ikey.user_key)) {
// For this user key:
// (1) there is no data in higher levels
// (2) data in lower levels will have larger sequence numbers
// (3) data in layers that are being compacted here and have
// smaller sequence numbers will be dropped in the next
// few iterations of this loop (by rule (A) above).
// Therefore this deletion marker is obsolete and can be dropped.
drop = true;
++key_drop_obsolete;
} else if (ikey.type == kTypeMerge) {
if (!merge.HasOperator()) {
LogToBuffer(log_buffer, "Options::merge_operator is null.");
status = Status::InvalidArgument(
"merge_operator is not properly initialized.");
break;
}
// We know the merge type entry is not hidden, otherwise we would
// have hit (A)
// We encapsulate the merge related state machine in a different
// object to minimize change to the existing flow. Turn out this
// logic could also be nicely re-used for memtable flush purge
// optimization in BuildTable.
int steps = 0;
merge.MergeUntil(input, prev_snapshot, bottommost_level,
db_options_.statistics.get(), &steps);
// Skip the Merge ops
combined_idx = combined_idx - 1 + steps;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
current_entry_is_merging = true;
if (merge.IsSuccess()) {
// Successfully found Put/Delete/(end-of-key-range) while merging
// Get the merge result
key = merge.key();
ParseInternalKey(key, &ikey);
value = merge.value();
} else {
// Did not find a Put/Delete/(end-of-key-range) while merging
// We now have some stack of merge operands to write out.
// NOTE: key,value, and ikey are now referring to old entries.
// These will be correctly set below.
assert(!merge.keys().empty());
assert(merge.keys().size() == merge.values().size());
// Hack to make sure last_sequence_for_key is correct
ParseInternalKey(merge.keys().front(), &ikey);
}
}
last_sequence_for_key = ikey.sequence;
visible_in_snapshot = visible;
}
if (!drop) {
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// We may write a single key (e.g.: for Put/Delete or successful merge).
// Or we may instead have to write a sequence/list of keys.
// We have to write a sequence iff we have an unsuccessful merge
bool has_merge_list = current_entry_is_merging && !merge.IsSuccess();
const std::deque<std::string>* keys = nullptr;
const std::deque<std::string>* values = nullptr;
std::deque<std::string>::const_reverse_iterator key_iter;
std::deque<std::string>::const_reverse_iterator value_iter;
if (has_merge_list) {
keys = &merge.keys();
values = &merge.values();
key_iter = keys->rbegin(); // The back (*rbegin()) is the first key
value_iter = values->rbegin();
key = Slice(*key_iter);
value = Slice(*value_iter);
}
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// If we have a list of keys to write, traverse the list.
// If we have a single key to write, simply write that key.
while (true) {
// Invariant: key,value,ikey will always be the next entry to write
char* kptr = (char*)key.data();
std::string kstr;
// Zeroing out the sequence number leads to better compression.
// If this is the bottommost level (no files in lower levels)
// and the earliest snapshot is larger than this seqno
// then we can squash the seqno to zero.
if (bottommost_level && ikey.sequence < earliest_snapshot &&
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
ikey.type != kTypeMerge) {
assert(ikey.type != kTypeDeletion);
// make a copy because updating in place would cause problems
// with the priority queue that is managing the input key iterator
kstr.assign(key.data(), key.size());
kptr = (char *)kstr.c_str();
UpdateInternalKey(kptr, key.size(), (uint64_t)0, ikey.type);
}
Slice newkey(kptr, key.size());
assert((key.clear(), 1)); // we do not need 'key' anymore
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// Open output file if necessary
if (compact->builder == nullptr) {
status = OpenCompactionOutputFile(compact, mutable_cf_options);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
if (!status.ok()) {
break;
}
}
SequenceNumber seqno = GetInternalKeySeqno(newkey);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
if (compact->builder->NumEntries() == 0) {
compact->current_output()->smallest.DecodeFrom(newkey);
compact->current_output()->smallest_seqno = seqno;
} else {
compact->current_output()->smallest_seqno =
std::min(compact->current_output()->smallest_seqno, seqno);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
}
compact->current_output()->largest.DecodeFrom(newkey);
compact->builder->Add(newkey, value);
(*num_output_records)++,
compact->current_output()->largest_seqno =
std::max(compact->current_output()->largest_seqno, seqno);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// Close output file if it is big enough
if (compact->builder->FileSize() >=
compact->compaction->MaxOutputFileSize()) {
status = FinishCompactionOutputFile(compact, input);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
if (!status.ok()) {
break;
}
}
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// If we have a list of entries, move to next element
// If we only had one entry, then break the loop.
if (has_merge_list) {
++key_iter;
++value_iter;
// If at end of list
if (key_iter == keys->rend() || value_iter == values->rend()) {
// Sanity Check: if one ends, then both end
assert(key_iter == keys->rend() && value_iter == values->rend());
break;
}
// Otherwise not at end of list. Update key, value, and ikey.
key = Slice(*key_iter);
value = Slice(*value_iter);
ParseInternalKey(key, &ikey);
} else{
// Only had one item to begin with (Put/Delete)
break;
}
} // while (true)
} // if (!drop)
// MergeUntil has moved input to the next entry
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
if (!current_entry_is_merging) {
input->Next();
}
}
if (key_drop_user > 0) {
RecordTick(stats_, COMPACTION_KEY_DROP_USER, key_drop_user);
}
if (key_drop_newer_entry > 0) {
RecordTick(stats_, COMPACTION_KEY_DROP_NEWER_ENTRY, key_drop_newer_entry);
}
if (key_drop_obsolete > 0) {
RecordTick(stats_, COMPACTION_KEY_DROP_OBSOLETE, key_drop_obsolete);
}
RecordCompactionIOStats();
return status;
}
void DBImpl::CallCompactionFilterV2(CompactionState* compact,
CompactionFilterV2* compaction_filter_v2) {
if (compact == nullptr || compaction_filter_v2 == nullptr) {
return;
}
// Assemble slice vectors for user keys and existing values.
// We also keep track of our parsed internal key structs because
// we may need to access the sequence number in the event that
// keys are garbage collected during the filter process.
std::vector<ParsedInternalKey> ikey_buf;
std::vector<Slice> user_key_buf;
std::vector<Slice> existing_value_buf;
for (const auto& key : compact->key_str_buf_) {
ParsedInternalKey ikey;
ParseInternalKey(Slice(key), &ikey);
ikey_buf.emplace_back(ikey);
user_key_buf.emplace_back(ikey.user_key);
}
for (const auto& value : compact->existing_value_str_buf_) {
existing_value_buf.emplace_back(Slice(value));
}
// If the user has specified a compaction filter and the sequence
// number is greater than any external snapshot, then invoke the
// filter.
// If the return value of the compaction filter is true, replace
// the entry with a delete marker.
compact->to_delete_buf_ = compaction_filter_v2->Filter(
compact->compaction->level(),
user_key_buf, existing_value_buf,
&compact->new_value_buf_,
&compact->value_changed_buf_);
// new_value_buf_.size() <= to_delete__buf_.size(). "=" iff all
// kv-pairs in this compaction run needs to be deleted.
assert(compact->to_delete_buf_.size() ==
compact->key_str_buf_.size());
assert(compact->to_delete_buf_.size() ==
compact->existing_value_str_buf_.size());
assert(compact->to_delete_buf_.size() ==
compact->value_changed_buf_.size());
int new_value_idx = 0;
for (unsigned int i = 0; i < compact->to_delete_buf_.size(); ++i) {
if (compact->to_delete_buf_[i]) {
// update the string buffer directly
// the Slice buffer points to the updated buffer
UpdateInternalKey(&compact->key_str_buf_[i][0],
compact->key_str_buf_[i].size(),
ikey_buf[i].sequence,
kTypeDeletion);
// no value associated with delete
compact->existing_value_str_buf_[i].clear();
RecordTick(stats_, COMPACTION_KEY_DROP_USER);
} else if (compact->value_changed_buf_[i]) {
compact->existing_value_str_buf_[i] =
compact->new_value_buf_[new_value_idx++];
}
} // for
}
Status DBImpl::DoCompactionWork(CompactionState* compact,
const MutableCFOptions& mutable_cf_options,
JobContext* job_context,
LogBuffer* log_buffer) {
assert(compact);
compact->CleanupBatchBuffer();
compact->CleanupMergedBuffer();
// Generate file_levels_ for compaction berfore making Iterator
compact->compaction->GenerateFileLevels();
int64_t imm_micros = 0; // Micros spent doing imm_ compactions
ColumnFamilyData* cfd = compact->compaction->column_family_data();
LogToBuffer(
log_buffer,
"[%s] Compacting %d@%d + %d@%d files, score %.2f slots available %d",
cfd->GetName().c_str(), compact->compaction->num_input_files(0),
compact->compaction->level(), compact->compaction->num_input_files(1),
compact->compaction->output_level(), compact->compaction->score(),
db_options_.max_background_compactions - bg_compaction_scheduled_);
char scratch[2345];
compact->compaction->Summary(scratch, sizeof(scratch));
LogToBuffer(log_buffer, "[%s] Compaction start summary: %s\n",
cfd->GetName().c_str(), scratch);
assert(cfd->current()->GetStorageInfo()->NumLevelFiles(
compact->compaction->level()) > 0);
assert(compact->builder == nullptr);
assert(!compact->outfile);
SequenceNumber visible_at_tip = 0;
SequenceNumber earliest_snapshot;
SequenceNumber latest_snapshot = 0;
snapshots_.getAll(compact->existing_snapshots);
if (compact->existing_snapshots.size() == 0) {
// optimize for fast path if there are no snapshots
visible_at_tip = versions_->LastSequence();
earliest_snapshot = visible_at_tip;
} else {
latest_snapshot = compact->existing_snapshots.back();
// Add the current seqno as the 'latest' virtual
// snapshot to the end of this list.
compact->existing_snapshots.push_back(versions_->LastSequence());
earliest_snapshot = compact->existing_snapshots[0];
}
// Is this compaction producing files at the bottommost level?
bool bottommost_level = compact->compaction->BottomMostLevel();
// Allocate the output file numbers before we release the lock
AllocateCompactionOutputFileNumbers(compact);
bool is_snapshot_supported = IsSnapshotSupported();
// Release mutex while we're actually doing the compaction work
mutex_.Unlock();
log_buffer->FlushBufferToLog();
int num_output_records = 0;
const uint64_t start_micros = env_->NowMicros();
unique_ptr<Iterator> input(versions_->MakeInputIterator(compact->compaction));
input->SeekToFirst();
Status status;
ParsedInternalKey ikey;
std::unique_ptr<CompactionFilterV2> compaction_filter_from_factory_v2
= nullptr;
auto context = compact->GetFilterContext();
compaction_filter_from_factory_v2 =
cfd->ioptions()->compaction_filter_factory_v2->
CreateCompactionFilterV2(context);
auto compaction_filter_v2 =
compaction_filter_from_factory_v2.get();
if (!compaction_filter_v2) {
status = ProcessKeyValueCompaction(
mutable_cf_options, is_snapshot_supported, visible_at_tip,
earliest_snapshot, latest_snapshot, job_context, bottommost_level,
&imm_micros, input.get(), compact, false, &num_output_records,
log_buffer);
} else {
// temp_backup_input always point to the start of the current buffer
// temp_backup_input = backup_input;
// iterate through input,
// 1) buffer ineligible keys and value keys into 2 separate buffers;
// 2) send value_buffer to compaction filter and alternate the values;
// 3) merge value_buffer with ineligible_value_buffer;
// 4) run the modified "compaction" using the old for loop.
bool prefix_initialized = false;
shared_ptr<Iterator> backup_input(
versions_->MakeInputIterator(compact->compaction));
backup_input->SeekToFirst();
while (backup_input->Valid() &&
!shutting_down_.load(std::memory_order_acquire) &&
!cfd->IsDropped()) {
// FLUSH preempts compaction
// TODO(icanadi) this currently only checks if flush is necessary on
// compacting column family. we should also check if flush is necessary on
// other column families, too
imm_micros += CallFlushDuringCompaction(cfd, mutable_cf_options,
job_context, log_buffer);
Slice key = backup_input->key();
Slice value = backup_input->value();
if (!ParseInternalKey(key, &ikey)) {
// log error
Log(db_options_.info_log, "[%s] Failed to parse key: %s",
cfd->GetName().c_str(), key.ToString().c_str());
continue;
} else {
const SliceTransform* transformer =
cfd->ioptions()->compaction_filter_factory_v2->GetPrefixExtractor();
const auto key_prefix = transformer->Transform(ikey.user_key);
if (!prefix_initialized) {
compact->cur_prefix_ = key_prefix.ToString();
prefix_initialized = true;
}
// If the prefix remains the same, keep buffering
if (key_prefix.compare(Slice(compact->cur_prefix_)) == 0) {
// Apply the compaction filter V2 to all the kv pairs sharing
// the same prefix
if (ikey.type == kTypeValue &&
(visible_at_tip || ikey.sequence > latest_snapshot)) {
// Buffer all keys sharing the same prefix for CompactionFilterV2
// Iterate through keys to check prefix
compact->BufferKeyValueSlices(key, value);
} else {
// buffer ineligible keys
compact->BufferOtherKeyValueSlices(key, value);
}
backup_input->Next();
continue;
// finish changing values for eligible keys
} else {
// Now prefix changes, this batch is done.
// Call compaction filter on the buffered values to change the value
if (compact->key_str_buf_.size() > 0) {
CallCompactionFilterV2(compact, compaction_filter_v2);
}
compact->cur_prefix_ = key_prefix.ToString();
}
}
// Merge this batch of data (values + ineligible keys)
compact->MergeKeyValueSliceBuffer(&cfd->internal_comparator());
// Done buffering for the current prefix. Spit it out to disk
// Now just iterate through all the kv-pairs
status = ProcessKeyValueCompaction(
mutable_cf_options, is_snapshot_supported, visible_at_tip,
earliest_snapshot, latest_snapshot, job_context, bottommost_level,
&imm_micros, input.get(), compact, true, &num_output_records,
log_buffer);
if (!status.ok()) {
break;
}
// After writing the kv-pairs, we can safely remove the reference
// to the string buffer and clean them up
compact->CleanupBatchBuffer();
compact->CleanupMergedBuffer();
// Buffer the key that triggers the mismatch in prefix
if (ikey.type == kTypeValue &&
(visible_at_tip || ikey.sequence > latest_snapshot)) {
compact->BufferKeyValueSlices(key, value);
} else {
compact->BufferOtherKeyValueSlices(key, value);
}
backup_input->Next();
if (!backup_input->Valid()) {
// If this is the single last value, we need to merge it.
if (compact->key_str_buf_.size() > 0) {
CallCompactionFilterV2(compact, compaction_filter_v2);
}
compact->MergeKeyValueSliceBuffer(&cfd->internal_comparator());
status = ProcessKeyValueCompaction(
mutable_cf_options, is_snapshot_supported, visible_at_tip,
earliest_snapshot, latest_snapshot, job_context, bottommost_level,
&imm_micros, input.get(), compact, true, &num_output_records,
log_buffer);
compact->CleanupBatchBuffer();
compact->CleanupMergedBuffer();
}
} // done processing all prefix batches
// finish the last batch
if (compact->key_str_buf_.size() > 0) {
CallCompactionFilterV2(compact, compaction_filter_v2);
}
compact->MergeKeyValueSliceBuffer(&cfd->internal_comparator());
status = ProcessKeyValueCompaction(
mutable_cf_options, is_snapshot_supported, visible_at_tip,
earliest_snapshot, latest_snapshot, job_context, bottommost_level,
&imm_micros, input.get(), compact, true, &num_output_records,
log_buffer);
} // checking for compaction filter v2
if (status.ok() &&
(shutting_down_.load(std::memory_order_acquire) || cfd->IsDropped())) {
status = Status::ShutdownInProgress(
"Database shutdown or Column family drop during compaction");
}
if (status.ok() && compact->builder != nullptr) {
status = FinishCompactionOutputFile(compact, input.get());
}
if (status.ok()) {
status = input->status();
}
input.reset();
if (!db_options_.disableDataSync) {
db_directory_->Fsync();
}
InternalStats::CompactionStats stats(1);
stats.micros = env_->NowMicros() - start_micros - imm_micros;
Improve statistics Summary: This adds more statistics to be reported by GetProperty("leveldb.stats"). The new stats include time spent waiting on stalls in MakeRoomForWrite. This also includes the total amplification rate where that is: (#bytes of sequential IO during compaction) / (#bytes from Put) This also includes a lot more data for the per-level compaction report. * Rn(MB) - MB read from level N during compaction between levels N and N+1 * Rnp1(MB) - MB read from level N+1 during compaction between levels N and N+1 * Wnew(MB) - new data written to the level during compaction * Amplify - ( Write(MB) + Rnp1(MB) ) / Rn(MB) * Rn - files read from level N during compaction between levels N and N+1 * Rnp1 - files read from level N+1 during compaction between levels N and N+1 * Wnp1 - files written to level N+1 during compaction between levels N and N+1 * NewW - new files written to level N+1 during compaction * Count - number of compactions done for this level This is the new output from DB::GetProperty("leveldb.stats"). The old output stopped at Write(MB) Compactions Level Files Size(MB) Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count ------------------------------------------------------------------------------------------------------------------------------------- 0 3 6 33 0 576 0 0 576 -1.0 0.0 1.3 0 0 0 0 290 1 127 242 351 5316 5314 570 4747 567 17.0 12.1 12.1 287 2399 2685 286 32 2 161 328 54 822 824 326 496 328 4.0 1.9 1.9 160 251 411 160 161 Amplification: 22.3 rate, 0.56 GB in, 12.55 GB out Uptime(secs): 439.8 Stalls(secs): 206.938 level0_slowdown, 0.000 level0_numfiles, 24.129 memtable_compaction Task ID: # Blame Rev: Test Plan: run db_bench Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - (cherry picked from commit ecdeead38f86cc02e754d0032600742c4f02fec8) Reviewers: dhruba Differential Revision: https://reviews.facebook.net/D6153
12 years ago
stats.files_in_leveln = compact->compaction->num_input_files(0);
stats.files_in_levelnp1 = compact->compaction->num_input_files(1);
MeasureTime(stats_, COMPACTION_TIME, stats.micros);
int num_output_files = compact->outputs.size();
if (compact->builder != nullptr) {
11 years ago
// An error occurred so ignore the last output.
assert(num_output_files > 0);
--num_output_files;
}
stats.files_out_levelnp1 = num_output_files;
Improve statistics Summary: This adds more statistics to be reported by GetProperty("leveldb.stats"). The new stats include time spent waiting on stalls in MakeRoomForWrite. This also includes the total amplification rate where that is: (#bytes of sequential IO during compaction) / (#bytes from Put) This also includes a lot more data for the per-level compaction report. * Rn(MB) - MB read from level N during compaction between levels N and N+1 * Rnp1(MB) - MB read from level N+1 during compaction between levels N and N+1 * Wnew(MB) - new data written to the level during compaction * Amplify - ( Write(MB) + Rnp1(MB) ) / Rn(MB) * Rn - files read from level N during compaction between levels N and N+1 * Rnp1 - files read from level N+1 during compaction between levels N and N+1 * Wnp1 - files written to level N+1 during compaction between levels N and N+1 * NewW - new files written to level N+1 during compaction * Count - number of compactions done for this level This is the new output from DB::GetProperty("leveldb.stats"). The old output stopped at Write(MB) Compactions Level Files Size(MB) Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count ------------------------------------------------------------------------------------------------------------------------------------- 0 3 6 33 0 576 0 0 576 -1.0 0.0 1.3 0 0 0 0 290 1 127 242 351 5316 5314 570 4747 567 17.0 12.1 12.1 287 2399 2685 286 32 2 161 328 54 822 824 326 496 328 4.0 1.9 1.9 160 251 411 160 161 Amplification: 22.3 rate, 0.56 GB in, 12.55 GB out Uptime(secs): 439.8 Stalls(secs): 206.938 level0_slowdown, 0.000 level0_numfiles, 24.129 memtable_compaction Task ID: # Blame Rev: Test Plan: run db_bench Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - (cherry picked from commit ecdeead38f86cc02e754d0032600742c4f02fec8) Reviewers: dhruba Differential Revision: https://reviews.facebook.net/D6153
12 years ago
uint64_t num_input_records = 0;
Add monitoring for universal compaction and add counters for compaction IO Summary: Adds these counters { WAL_FILE_SYNCED, "rocksdb.wal.synced" } number of writes that request a WAL sync { WAL_FILE_BYTES, "rocksdb.wal.bytes" }, number of bytes written to the WAL { WRITE_DONE_BY_SELF, "rocksdb.write.self" }, number of writes processed by the calling thread { WRITE_DONE_BY_OTHER, "rocksdb.write.other" }, number of writes not processed by the calling thread. Instead these were processed by the current holder of the write lock { WRITE_WITH_WAL, "rocksdb.write.wal" }, number of writes that request WAL logging { COMPACT_READ_BYTES, "rocksdb.compact.read.bytes" }, number of bytes read during compaction { COMPACT_WRITE_BYTES, "rocksdb.compact.write.bytes" }, number of bytes written during compaction Per-interval stats output was updated with WAL stats and correct stats for universal compaction including a correct value for write-amplification. It now looks like: Compactions Level Files Size(MB) Score Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) RW-Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count Ln-stall Stall-cnt -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 0 7 464 46.4 281 3411 3875 3411 0 3875 2.1 12.1 13.8 621 0 240 240 628 0.0 0 Uptime(secs): 310.8 total, 2.0 interval Writes cumulative: 9999999 total, 9999999 batches, 1.0 per batch, 1.22 ingest GB WAL cumulative: 9999999 WAL writes, 9999999 WAL syncs, 1.00 writes per sync, 1.22 GB written Compaction IO cumulative (GB): 1.22 new, 3.33 read, 3.78 write, 7.12 read+write Compaction IO cumulative (MB/sec): 4.0 new, 11.0 read, 12.5 write, 23.4 read+write Amplification cumulative: 4.1 write, 6.8 compaction Writes interval: 100000 total, 100000 batches, 1.0 per batch, 12.5 ingest MB WAL interval: 100000 WAL writes, 100000 WAL syncs, 1.00 writes per sync, 0.01 MB written Compaction IO interval (MB): 12.49 new, 14.98 read, 21.50 write, 36.48 read+write Compaction IO interval (MB/sec): 6.4 new, 7.6 read, 11.0 write, 18.6 read+write Amplification interval: 101.7 write, 102.9 compaction Stalls(secs): 142.924 level0_slowdown, 0.000 level0_numfiles, 0.805 memtable_compaction, 0.000 leveln_slowdown Stalls(count): 132461 level0_slowdown, 0 level0_numfiles, 3 memtable_compaction, 0 leveln_slowdown Task ID: #3329644, #3301695 Blame Rev: Test Plan: Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - Reviewers: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D14583
11 years ago
for (int i = 0; i < compact->compaction->num_input_files(0); i++) {
stats.bytes_readn += compact->compaction->input(0, i)->fd.GetFileSize();
stats.num_input_records += compact->compaction->input(0, i)->num_entries;
num_input_records += compact->compaction->input(0, i)->num_entries;
Add monitoring for universal compaction and add counters for compaction IO Summary: Adds these counters { WAL_FILE_SYNCED, "rocksdb.wal.synced" } number of writes that request a WAL sync { WAL_FILE_BYTES, "rocksdb.wal.bytes" }, number of bytes written to the WAL { WRITE_DONE_BY_SELF, "rocksdb.write.self" }, number of writes processed by the calling thread { WRITE_DONE_BY_OTHER, "rocksdb.write.other" }, number of writes not processed by the calling thread. Instead these were processed by the current holder of the write lock { WRITE_WITH_WAL, "rocksdb.write.wal" }, number of writes that request WAL logging { COMPACT_READ_BYTES, "rocksdb.compact.read.bytes" }, number of bytes read during compaction { COMPACT_WRITE_BYTES, "rocksdb.compact.write.bytes" }, number of bytes written during compaction Per-interval stats output was updated with WAL stats and correct stats for universal compaction including a correct value for write-amplification. It now looks like: Compactions Level Files Size(MB) Score Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) RW-Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count Ln-stall Stall-cnt -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 0 7 464 46.4 281 3411 3875 3411 0 3875 2.1 12.1 13.8 621 0 240 240 628 0.0 0 Uptime(secs): 310.8 total, 2.0 interval Writes cumulative: 9999999 total, 9999999 batches, 1.0 per batch, 1.22 ingest GB WAL cumulative: 9999999 WAL writes, 9999999 WAL syncs, 1.00 writes per sync, 1.22 GB written Compaction IO cumulative (GB): 1.22 new, 3.33 read, 3.78 write, 7.12 read+write Compaction IO cumulative (MB/sec): 4.0 new, 11.0 read, 12.5 write, 23.4 read+write Amplification cumulative: 4.1 write, 6.8 compaction Writes interval: 100000 total, 100000 batches, 1.0 per batch, 12.5 ingest MB WAL interval: 100000 WAL writes, 100000 WAL syncs, 1.00 writes per sync, 0.01 MB written Compaction IO interval (MB): 12.49 new, 14.98 read, 21.50 write, 36.48 read+write Compaction IO interval (MB/sec): 6.4 new, 7.6 read, 11.0 write, 18.6 read+write Amplification interval: 101.7 write, 102.9 compaction Stalls(secs): 142.924 level0_slowdown, 0.000 level0_numfiles, 0.805 memtable_compaction, 0.000 leveln_slowdown Stalls(count): 132461 level0_slowdown, 0 level0_numfiles, 3 memtable_compaction, 0 leveln_slowdown Task ID: #3329644, #3301695 Blame Rev: Test Plan: Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - Reviewers: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D14583
11 years ago
}
Improve statistics Summary: This adds more statistics to be reported by GetProperty("leveldb.stats"). The new stats include time spent waiting on stalls in MakeRoomForWrite. This also includes the total amplification rate where that is: (#bytes of sequential IO during compaction) / (#bytes from Put) This also includes a lot more data for the per-level compaction report. * Rn(MB) - MB read from level N during compaction between levels N and N+1 * Rnp1(MB) - MB read from level N+1 during compaction between levels N and N+1 * Wnew(MB) - new data written to the level during compaction * Amplify - ( Write(MB) + Rnp1(MB) ) / Rn(MB) * Rn - files read from level N during compaction between levels N and N+1 * Rnp1 - files read from level N+1 during compaction between levels N and N+1 * Wnp1 - files written to level N+1 during compaction between levels N and N+1 * NewW - new files written to level N+1 during compaction * Count - number of compactions done for this level This is the new output from DB::GetProperty("leveldb.stats"). The old output stopped at Write(MB) Compactions Level Files Size(MB) Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count ------------------------------------------------------------------------------------------------------------------------------------- 0 3 6 33 0 576 0 0 576 -1.0 0.0 1.3 0 0 0 0 290 1 127 242 351 5316 5314 570 4747 567 17.0 12.1 12.1 287 2399 2685 286 32 2 161 328 54 822 824 326 496 328 4.0 1.9 1.9 160 251 411 160 161 Amplification: 22.3 rate, 0.56 GB in, 12.55 GB out Uptime(secs): 439.8 Stalls(secs): 206.938 level0_slowdown, 0.000 level0_numfiles, 24.129 memtable_compaction Task ID: # Blame Rev: Test Plan: run db_bench Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - (cherry picked from commit ecdeead38f86cc02e754d0032600742c4f02fec8) Reviewers: dhruba Differential Revision: https://reviews.facebook.net/D6153
12 years ago
Add monitoring for universal compaction and add counters for compaction IO Summary: Adds these counters { WAL_FILE_SYNCED, "rocksdb.wal.synced" } number of writes that request a WAL sync { WAL_FILE_BYTES, "rocksdb.wal.bytes" }, number of bytes written to the WAL { WRITE_DONE_BY_SELF, "rocksdb.write.self" }, number of writes processed by the calling thread { WRITE_DONE_BY_OTHER, "rocksdb.write.other" }, number of writes not processed by the calling thread. Instead these were processed by the current holder of the write lock { WRITE_WITH_WAL, "rocksdb.write.wal" }, number of writes that request WAL logging { COMPACT_READ_BYTES, "rocksdb.compact.read.bytes" }, number of bytes read during compaction { COMPACT_WRITE_BYTES, "rocksdb.compact.write.bytes" }, number of bytes written during compaction Per-interval stats output was updated with WAL stats and correct stats for universal compaction including a correct value for write-amplification. It now looks like: Compactions Level Files Size(MB) Score Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) RW-Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count Ln-stall Stall-cnt -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 0 7 464 46.4 281 3411 3875 3411 0 3875 2.1 12.1 13.8 621 0 240 240 628 0.0 0 Uptime(secs): 310.8 total, 2.0 interval Writes cumulative: 9999999 total, 9999999 batches, 1.0 per batch, 1.22 ingest GB WAL cumulative: 9999999 WAL writes, 9999999 WAL syncs, 1.00 writes per sync, 1.22 GB written Compaction IO cumulative (GB): 1.22 new, 3.33 read, 3.78 write, 7.12 read+write Compaction IO cumulative (MB/sec): 4.0 new, 11.0 read, 12.5 write, 23.4 read+write Amplification cumulative: 4.1 write, 6.8 compaction Writes interval: 100000 total, 100000 batches, 1.0 per batch, 12.5 ingest MB WAL interval: 100000 WAL writes, 100000 WAL syncs, 1.00 writes per sync, 0.01 MB written Compaction IO interval (MB): 12.49 new, 14.98 read, 21.50 write, 36.48 read+write Compaction IO interval (MB/sec): 6.4 new, 7.6 read, 11.0 write, 18.6 read+write Amplification interval: 101.7 write, 102.9 compaction Stalls(secs): 142.924 level0_slowdown, 0.000 level0_numfiles, 0.805 memtable_compaction, 0.000 leveln_slowdown Stalls(count): 132461 level0_slowdown, 0 level0_numfiles, 3 memtable_compaction, 0 leveln_slowdown Task ID: #3329644, #3301695 Blame Rev: Test Plan: Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - Reviewers: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D14583
11 years ago
for (int i = 0; i < compact->compaction->num_input_files(1); i++) {
stats.bytes_readnp1 += compact->compaction->input(1, i)->fd.GetFileSize();
num_input_records += compact->compaction->input(1, i)->num_entries;
Add monitoring for universal compaction and add counters for compaction IO Summary: Adds these counters { WAL_FILE_SYNCED, "rocksdb.wal.synced" } number of writes that request a WAL sync { WAL_FILE_BYTES, "rocksdb.wal.bytes" }, number of bytes written to the WAL { WRITE_DONE_BY_SELF, "rocksdb.write.self" }, number of writes processed by the calling thread { WRITE_DONE_BY_OTHER, "rocksdb.write.other" }, number of writes not processed by the calling thread. Instead these were processed by the current holder of the write lock { WRITE_WITH_WAL, "rocksdb.write.wal" }, number of writes that request WAL logging { COMPACT_READ_BYTES, "rocksdb.compact.read.bytes" }, number of bytes read during compaction { COMPACT_WRITE_BYTES, "rocksdb.compact.write.bytes" }, number of bytes written during compaction Per-interval stats output was updated with WAL stats and correct stats for universal compaction including a correct value for write-amplification. It now looks like: Compactions Level Files Size(MB) Score Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) RW-Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count Ln-stall Stall-cnt -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 0 7 464 46.4 281 3411 3875 3411 0 3875 2.1 12.1 13.8 621 0 240 240 628 0.0 0 Uptime(secs): 310.8 total, 2.0 interval Writes cumulative: 9999999 total, 9999999 batches, 1.0 per batch, 1.22 ingest GB WAL cumulative: 9999999 WAL writes, 9999999 WAL syncs, 1.00 writes per sync, 1.22 GB written Compaction IO cumulative (GB): 1.22 new, 3.33 read, 3.78 write, 7.12 read+write Compaction IO cumulative (MB/sec): 4.0 new, 11.0 read, 12.5 write, 23.4 read+write Amplification cumulative: 4.1 write, 6.8 compaction Writes interval: 100000 total, 100000 batches, 1.0 per batch, 12.5 ingest MB WAL interval: 100000 WAL writes, 100000 WAL syncs, 1.00 writes per sync, 0.01 MB written Compaction IO interval (MB): 12.49 new, 14.98 read, 21.50 write, 36.48 read+write Compaction IO interval (MB/sec): 6.4 new, 7.6 read, 11.0 write, 18.6 read+write Amplification interval: 101.7 write, 102.9 compaction Stalls(secs): 142.924 level0_slowdown, 0.000 level0_numfiles, 0.805 memtable_compaction, 0.000 leveln_slowdown Stalls(count): 132461 level0_slowdown, 0 level0_numfiles, 3 memtable_compaction, 0 leveln_slowdown Task ID: #3329644, #3301695 Blame Rev: Test Plan: Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - Reviewers: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D14583
11 years ago
}
Improve statistics Summary: This adds more statistics to be reported by GetProperty("leveldb.stats"). The new stats include time spent waiting on stalls in MakeRoomForWrite. This also includes the total amplification rate where that is: (#bytes of sequential IO during compaction) / (#bytes from Put) This also includes a lot more data for the per-level compaction report. * Rn(MB) - MB read from level N during compaction between levels N and N+1 * Rnp1(MB) - MB read from level N+1 during compaction between levels N and N+1 * Wnew(MB) - new data written to the level during compaction * Amplify - ( Write(MB) + Rnp1(MB) ) / Rn(MB) * Rn - files read from level N during compaction between levels N and N+1 * Rnp1 - files read from level N+1 during compaction between levels N and N+1 * Wnp1 - files written to level N+1 during compaction between levels N and N+1 * NewW - new files written to level N+1 during compaction * Count - number of compactions done for this level This is the new output from DB::GetProperty("leveldb.stats"). The old output stopped at Write(MB) Compactions Level Files Size(MB) Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count ------------------------------------------------------------------------------------------------------------------------------------- 0 3 6 33 0 576 0 0 576 -1.0 0.0 1.3 0 0 0 0 290 1 127 242 351 5316 5314 570 4747 567 17.0 12.1 12.1 287 2399 2685 286 32 2 161 328 54 822 824 326 496 328 4.0 1.9 1.9 160 251 411 160 161 Amplification: 22.3 rate, 0.56 GB in, 12.55 GB out Uptime(secs): 439.8 Stalls(secs): 206.938 level0_slowdown, 0.000 level0_numfiles, 24.129 memtable_compaction Task ID: # Blame Rev: Test Plan: run db_bench Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - (cherry picked from commit ecdeead38f86cc02e754d0032600742c4f02fec8) Reviewers: dhruba Differential Revision: https://reviews.facebook.net/D6153
12 years ago
for (int i = 0; i < num_output_files; i++) {
stats.bytes_written += compact->outputs[i].file_size;
}
stats.num_dropped_records =
static_cast<int>(num_input_records) - num_output_records;
RecordCompactionIOStats();
LogFlush(db_options_.info_log);
mutex_.Lock();
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
cfd->internal_stats()->AddCompactionStats(
compact->compaction->output_level(), stats);
// if there were any unused file number (mostly in case of
// compaction error), free up the entry from pending_putputs
ReleaseCompactionUnusedFileNumbers(compact);
if (status.ok()) {
status = InstallCompactionResults(compact, mutable_cf_options, log_buffer);
InstallSuperVersionBackground(cfd, job_context, mutable_cf_options);
}
VersionStorageInfo::LevelSummaryStorage tmp;
LogToBuffer(log_buffer,
"[%s] compacted to: %s, MB/sec: %.1f rd, %.1f wr, level %d, "
"files in(%d, %d) out(%d) "
"MB in(%.1f, %.1f) out(%.1f), read-write-amplify(%.1f) "
"write-amplify(%.1f) %s, records in: %d, records dropped: %d\n",
cfd->GetName().c_str(),
cfd->current()->GetStorageInfo()->LevelSummary(&tmp),
(stats.bytes_readn + stats.bytes_readnp1) /
static_cast<double>(stats.micros),
stats.bytes_written / static_cast<double>(stats.micros),
compact->compaction->output_level(), stats.files_in_leveln,
stats.files_in_levelnp1, stats.files_out_levelnp1,
stats.bytes_readn / 1048576.0, stats.bytes_readnp1 / 1048576.0,
stats.bytes_written / 1048576.0,
(stats.bytes_written + stats.bytes_readnp1 + stats.bytes_readn) /
(double)stats.bytes_readn,
stats.bytes_written / (double)stats.bytes_readn,
status.ToString().c_str(), stats.num_input_records,
stats.num_dropped_records);
Improve statistics Summary: This adds more statistics to be reported by GetProperty("leveldb.stats"). The new stats include time spent waiting on stalls in MakeRoomForWrite. This also includes the total amplification rate where that is: (#bytes of sequential IO during compaction) / (#bytes from Put) This also includes a lot more data for the per-level compaction report. * Rn(MB) - MB read from level N during compaction between levels N and N+1 * Rnp1(MB) - MB read from level N+1 during compaction between levels N and N+1 * Wnew(MB) - new data written to the level during compaction * Amplify - ( Write(MB) + Rnp1(MB) ) / Rn(MB) * Rn - files read from level N during compaction between levels N and N+1 * Rnp1 - files read from level N+1 during compaction between levels N and N+1 * Wnp1 - files written to level N+1 during compaction between levels N and N+1 * NewW - new files written to level N+1 during compaction * Count - number of compactions done for this level This is the new output from DB::GetProperty("leveldb.stats"). The old output stopped at Write(MB) Compactions Level Files Size(MB) Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count ------------------------------------------------------------------------------------------------------------------------------------- 0 3 6 33 0 576 0 0 576 -1.0 0.0 1.3 0 0 0 0 290 1 127 242 351 5316 5314 570 4747 567 17.0 12.1 12.1 287 2399 2685 286 32 2 161 328 54 822 824 326 496 328 4.0 1.9 1.9 160 251 411 160 161 Amplification: 22.3 rate, 0.56 GB in, 12.55 GB out Uptime(secs): 439.8 Stalls(secs): 206.938 level0_slowdown, 0.000 level0_numfiles, 24.129 memtable_compaction Task ID: # Blame Rev: Test Plan: run db_bench Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - (cherry picked from commit ecdeead38f86cc02e754d0032600742c4f02fec8) Reviewers: dhruba Differential Revision: https://reviews.facebook.net/D6153
12 years ago
return status;
}
namespace {
struct IterState {
IterState(DBImpl* db, port::Mutex* mu, SuperVersion* super_version)
: db(db), mu(mu), super_version(super_version) {}
DBImpl* db;
port::Mutex* mu;
SuperVersion* super_version;
};
static void CleanupIteratorState(void* arg1, void* arg2) {
IterState* state = reinterpret_cast<IterState*>(arg1);
if (state->super_version->Unref()) {
JobContext job_context;
state->mu->Lock();
state->super_version->Cleanup();
state->db->FindObsoleteFiles(&job_context, false, true);
state->mu->Unlock();
delete state->super_version;
if (job_context.HaveSomethingToDelete()) {
state->db->PurgeObsoleteFiles(job_context);
}
MemTableListVersion Summary: MemTableListVersion is to MemTableList what Version is to VersionSet. I took almost the same ideas to develop MemTableListVersion. The reason is to have copying std::list done in background, while flushing, rather than in foreground (MultiGet() and NewIterator()) under a mutex! Also, whenever we copied MemTableList, we copied also some MemTableList metadata (flush_requested_, commit_in_progress_, etc.), which was wasteful. This diff avoids std::list copy under a mutex in both MultiGet() and NewIterator(). I created a small database with some number of immutable memtables, and creating 100.000 iterators in a single-thread (!) decreased from {188739, 215703, 198028} to {154352, 164035, 159817}. A lot of the savings come from code under a mutex, so we should see much higher savings with multiple threads. Creating new iterator is very important to LogDevice team. I also think this diff will make SuperVersion obsolete for performance reasons. I will try it in the next diff. SuperVersion gave us huge savings on Get() code path, but I think that most of the savings came from copying MemTableList under a mutex. If we had MemTableListVersion, we would never need to copy the entire object (like we still do in NewIterator() and MultiGet()) Test Plan: `make check` works. I will also do `make valgrind_check` before commit Reviewers: dhruba, haobo, kailiu, sdong, emayanke, tnovak Reviewed By: kailiu CC: leveldb Differential Revision: https://reviews.facebook.net/D15255
11 years ago
}
delete state;
}
} // namespace
Iterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
ColumnFamilyData* cfd,
SuperVersion* super_version,
Arena* arena) {
Iterator* internal_iter;
assert(arena != nullptr);
// Need to create internal iterator from the arena.
MergeIteratorBuilder merge_iter_builder(&cfd->internal_comparator(), arena);
// Collect iterator for mutable mem
merge_iter_builder.AddIterator(
super_version->mem->NewIterator(read_options, arena));
// Collect all needed child iterators for immutable memtables
super_version->imm->AddIterators(read_options, &merge_iter_builder);
// Collect iterators for files in L0 - Ln
super_version->current->AddIterators(read_options, env_options_,
&merge_iter_builder);
internal_iter = merge_iter_builder.Finish();
IterState* cleanup = new IterState(this, &mutex_, super_version);
internal_iter->RegisterCleanup(CleanupIteratorState, cleanup, nullptr);
return internal_iter;
}
ColumnFamilyHandle* DBImpl::DefaultColumnFamily() const {
return default_cf_handle_;
}
Status DBImpl::Get(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const Slice& key,
std::string* value) {
return GetImpl(read_options, column_family, key, value);
}
// JobContext gets created and destructed outside of the lock --
// we
// use this convinently to:
// * malloc one SuperVersion() outside of the lock -- new_superversion
// * delete SuperVersion()s outside of the lock -- superversions_to_free
//
// However, if InstallSuperVersion() gets called twice with the same
// job_context, we can't reuse the SuperVersion() that got
// malloced
// because
// first call already used it. In that rare case, we take a hit and create a
// new SuperVersion() inside of the mutex. We do similar thing
// for superversion_to_free
void DBImpl::InstallSuperVersionBackground(
ColumnFamilyData* cfd, JobContext* job_context,
const MutableCFOptions& mutable_cf_options) {
mutex_.AssertHeld();
SuperVersion* old_superversion = InstallSuperVersion(
cfd, job_context->new_superversion, mutable_cf_options);
job_context->new_superversion = nullptr;
job_context->superversions_to_free.push_back(old_superversion);
}
SuperVersion* DBImpl::InstallSuperVersion(
ColumnFamilyData* cfd, SuperVersion* new_sv,
const MutableCFOptions& mutable_cf_options) {
mutex_.AssertHeld();
auto* old = cfd->InstallSuperVersion(
new_sv ? new_sv : new SuperVersion(), &mutex_, mutable_cf_options);
// We want to schedule potential flush or compactions since new options may
// have been picked up in this new version. New options may cause flush
// compaction trigger condition to change.
MaybeScheduleFlushOrCompaction();
// Update max_total_in_memory_state_
auto old_memtable_size = 0;
if (old) {
old_memtable_size = old->mutable_cf_options.write_buffer_size *
old->mutable_cf_options.max_write_buffer_number;
}
max_total_in_memory_state_ =
max_total_in_memory_state_ - old_memtable_size +
mutable_cf_options.write_buffer_size *
mutable_cf_options.max_write_buffer_number;
return old;
}
Status DBImpl::GetImpl(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const Slice& key,
std::string* value, bool* value_found) {
StopWatch sw(env_, stats_, DB_GET);
PERF_TIMER_GUARD(get_snapshot_time);
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
SequenceNumber snapshot;
if (read_options.snapshot != nullptr) {
snapshot = reinterpret_cast<const SnapshotImpl*>(
read_options.snapshot)->number_;
} else {
snapshot = versions_->LastSequence();
}
// Acquire SuperVersion
SuperVersion* sv = GetAndRefSuperVersion(cfd);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// Prepare to store a list of merge operations if merge occurs.
MergeContext merge_context;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
Status s;
// First look in the memtable, then in the immutable memtable (if any).
// s is both in/out. When in, s could either be OK or MergeInProgress.
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// merge_operands will contain the sequence of merges in the latter case.
LookupKey lkey(key, snapshot);
PERF_TIMER_STOP(get_snapshot_time);
if (sv->mem->Get(lkey, value, &s, &merge_context)) {
// Done
RecordTick(stats_, MEMTABLE_HIT);
} else if (sv->imm->Get(lkey, value, &s, &merge_context)) {
// Done
RecordTick(stats_, MEMTABLE_HIT);
} else {
PERF_TIMER_GUARD(get_from_output_files_time);
sv->current->Get(read_options, lkey, value, &s, &merge_context,
value_found);
RecordTick(stats_, MEMTABLE_MISS);
}
{
PERF_TIMER_GUARD(get_post_process_time);
ReturnAndCleanupSuperVersion(cfd, sv);
RecordTick(stats_, NUMBER_KEYS_READ);
RecordTick(stats_, BYTES_READ, value->size());
}
return s;
}
std::vector<Status> DBImpl::MultiGet(
const ReadOptions& read_options,
const std::vector<ColumnFamilyHandle*>& column_family,
const std::vector<Slice>& keys, std::vector<std::string>* values) {
StopWatch sw(env_, stats_, DB_MULTIGET);
PERF_TIMER_GUARD(get_snapshot_time);
SequenceNumber snapshot;
struct MultiGetColumnFamilyData {
ColumnFamilyData* cfd;
SuperVersion* super_version;
};
std::unordered_map<uint32_t, MultiGetColumnFamilyData*> multiget_cf_data;
// fill up and allocate outside of mutex
for (auto cf : column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(cf);
auto cfd = cfh->cfd();
if (multiget_cf_data.find(cfd->GetID()) == multiget_cf_data.end()) {
auto mgcfd = new MultiGetColumnFamilyData();
mgcfd->cfd = cfd;
multiget_cf_data.insert({cfd->GetID(), mgcfd});
}
}
mutex_.Lock();
if (read_options.snapshot != nullptr) {
snapshot = reinterpret_cast<const SnapshotImpl*>(
read_options.snapshot)->number_;
} else {
snapshot = versions_->LastSequence();
}
for (auto mgd_iter : multiget_cf_data) {
mgd_iter.second->super_version =
mgd_iter.second->cfd->GetSuperVersion()->Ref();
}
mutex_.Unlock();
// Contain a list of merge operations if merge occurs.
MergeContext merge_context;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// Note: this always resizes the values array
size_t num_keys = keys.size();
std::vector<Status> stat_list(num_keys);
values->resize(num_keys);
// Keep track of bytes that we read for statistics-recording later
uint64_t bytes_read = 0;
PERF_TIMER_STOP(get_snapshot_time);
// For each of the given keys, apply the entire "get" process as follows:
// First look in the memtable, then in the immutable memtable (if any).
// s is both in/out. When in, s could either be OK or MergeInProgress.
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
11 years ago
// merge_operands will contain the sequence of merges in the latter case.
for (size_t i = 0; i < num_keys; ++i) {
merge_context.Clear();
Status& s = stat_list[i];
std::string* value = &(*values)[i];
LookupKey lkey(keys[i], snapshot);
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family[i]);
auto mgd_iter = multiget_cf_data.find(cfh->cfd()->GetID());
assert(mgd_iter != multiget_cf_data.end());
auto mgd = mgd_iter->second;
auto super_version = mgd->super_version;
if (super_version->mem->Get(lkey, value, &s, &merge_context)) {
// Done
} else if (super_version->imm->Get(lkey, value, &s, &merge_context)) {
// Done
} else {
PERF_TIMER_GUARD(get_from_output_files_time);
super_version->current->Get(read_options, lkey, value, &s,
&merge_context);
}
if (s.ok()) {
bytes_read += value->size();
}
}
// Post processing (decrement reference counts and record statistics)
PERF_TIMER_GUARD(get_post_process_time);
autovector<SuperVersion*> superversions_to_delete;
// TODO(icanadi) do we need lock here or just around Cleanup()?
mutex_.Lock();
for (auto mgd_iter : multiget_cf_data) {
auto mgd = mgd_iter.second;
if (mgd->super_version->Unref()) {
mgd->super_version->Cleanup();
superversions_to_delete.push_back(mgd->super_version);
}
}
mutex_.Unlock();
for (auto td : superversions_to_delete) {
delete td;
}
for (auto mgd : multiget_cf_data) {
delete mgd.second;
}
RecordTick(stats_, NUMBER_MULTIGET_CALLS);
RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
RecordTick(stats_, NUMBER_MULTIGET_BYTES_READ, bytes_read);
PERF_TIMER_STOP(get_post_process_time);
return stat_list;
}
Status DBImpl::CreateColumnFamily(const ColumnFamilyOptions& cf_options,
const std::string& column_family_name,
ColumnFamilyHandle** handle) {
*handle = nullptr;
MutexLock l(&mutex_);
if (versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name) !=
nullptr) {
return Status::InvalidArgument("Column family already exists");
}
VersionEdit edit;
edit.AddColumnFamily(column_family_name);
uint32_t new_id = versions_->GetColumnFamilySet()->GetNextColumnFamilyID();
edit.SetColumnFamily(new_id);
edit.SetLogNumber(logfile_number_);
edit.SetComparatorName(cf_options.comparator->Name());
// LogAndApply will both write the creation in MANIFEST and create
// ColumnFamilyData object
Options opt(db_options_, cf_options);
Status s = versions_->LogAndApply(nullptr,
MutableCFOptions(opt, ImmutableCFOptions(opt)),
&edit, &mutex_, db_directory_.get(), false, &cf_options);
if (s.ok()) {
single_column_family_mode_ = false;
auto cfd =
versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name);
assert(cfd != nullptr);
delete InstallSuperVersion(cfd, nullptr, *cfd->GetLatestMutableCFOptions());
*handle = new ColumnFamilyHandleImpl(cfd, this, &mutex_);
Log(db_options_.info_log, "Created column family [%s] (ID %u)",
column_family_name.c_str(), (unsigned)cfd->GetID());
} else {
Log(db_options_.info_log, "Creating column family [%s] FAILED -- %s",
column_family_name.c_str(), s.ToString().c_str());
}
return s;
}
Status DBImpl::DropColumnFamily(ColumnFamilyHandle* column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
if (cfd->GetID() == 0) {
return Status::InvalidArgument("Can't drop default column family");
}
VersionEdit edit;
edit.DropColumnFamily();
edit.SetColumnFamily(cfd->GetID());
Status s;
{
MutexLock l(&mutex_);
if (cfd->IsDropped()) {
s = Status::InvalidArgument("Column family already dropped!\n");
}
if (s.ok()) {
// we drop column family from a single write thread
WriteThread::Writer w(&mutex_);
s = write_thread_.EnterWriteThread(&w, 0);
assert(s.ok() && !w.done); // No timeout and nobody should do our job
s = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
&edit, &mutex_);
write_thread_.ExitWriteThread(&w, &w, s);
}
}
if (s.ok()) {
assert(cfd->IsDropped());
auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
max_total_in_memory_state_ -= mutable_cf_options->write_buffer_size *
mutable_cf_options->max_write_buffer_number;
Log(db_options_.info_log, "Dropped column family with id %u\n",
cfd->GetID());
} else {
Log(db_options_.info_log,
"Dropping column family with id %u FAILED -- %s\n",
cfd->GetID(), s.ToString().c_str());
}
return s;
}
bool DBImpl::KeyMayExist(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const Slice& key,
std::string* value, bool* value_found) {
if (value_found != nullptr) {
// falsify later if key-may-exist but can't fetch value
*value_found = true;
}
ReadOptions roptions = read_options;
roptions.read_tier = kBlockCacheTier; // read from block cache only
auto s = GetImpl(roptions, column_family, key, value, value_found);
// If block_cache is enabled and the index block of the table didn't
// not present in block_cache, the return value will be Status::Incomplete.
// In this case, key may still exist in the table.
return s.ok() || s.IsIncomplete();
}
Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
ColumnFamilyHandle* column_family) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
if (read_options.tailing) {
#ifdef ROCKSDB_LITE
// not supported in lite version
return nullptr;
#else
SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
auto iter = new ForwardIterator(this, read_options, cfd, sv);
return NewDBIterator(env_, *cfd->ioptions(), cfd->user_comparator(), iter,
kMaxSequenceNumber,
sv->mutable_cf_options.max_sequential_skip_in_iterations,
read_options.iterate_upper_bound);
#endif
} else {
SequenceNumber latest_snapshot = versions_->LastSequence();
SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
auto snapshot =
read_options.snapshot != nullptr
? reinterpret_cast<const SnapshotImpl*>(
read_options.snapshot)->number_
: latest_snapshot;
// Try to generate a DB iterator tree in continuous memory area to be
// cache friendly. Here is an example of result:
// +-------------------------------+
// | |
// | ArenaWrappedDBIter |
// | + |
// | +---> Inner Iterator ------------+
// | | | |
// | | +-- -- -- -- -- -- -- --+ |
// | +--- | Arena | |
// | | | |
// | Allocated Memory: | |
// | | +-------------------+ |
// | | | DBIter | <---+
// | | + |
// | | | +-> iter_ ------------+
// | | | | |
// | | +-------------------+ |
// | | | MergingIterator | <---+
// | | + |
// | | | +->child iter1 ------------+
// | | | | | |
// | | +->child iter2 ----------+ |
// | | | | | | |
// | | | +->child iter3 --------+ | |
// | | | | | |
// | | +-------------------+ | | |
// | | | Iterator1 | <--------+
// | | +-------------------+ | |
// | | | Iterator2 | <------+
// | | +-------------------+ |
// | | | Iterator3 | <----+
// | | +-------------------+
// | | |
// +-------+-----------------------+
//
// ArenaWrappedDBIter inlines an arena area where all the iterartor in the
// the iterator tree is allocated in the order of being accessed when
// querying.
// Laying out the iterators in the order of being accessed makes it more
// likely that any iterator pointer is close to the iterator it points to so
// that they are likely to be in the same cache line and/or page.
ArenaWrappedDBIter* db_iter = NewArenaWrappedDbIterator(
env_, *cfd->ioptions(), cfd->user_comparator(),
snapshot, sv->mutable_cf_options.max_sequential_skip_in_iterations,
read_options.iterate_upper_bound);
Iterator* internal_iter =
NewInternalIterator(read_options, cfd, sv, db_iter->GetArena());
db_iter->SetIterUnderDBIter(internal_iter);
return db_iter;
}
// To stop compiler from complaining
return nullptr;
}
Status DBImpl::NewIterators(
const ReadOptions& read_options,
const std::vector<ColumnFamilyHandle*>& column_families,
std::vector<Iterator*>* iterators) {
iterators->clear();
iterators->reserve(column_families.size());
if (read_options.tailing) {
#ifdef ROCKSDB_LITE
return Status::InvalidArgument(
"Tailing interator not supported in RocksDB lite");
#else
for (auto cfh : column_families) {
auto cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(cfh)->cfd();
SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
auto iter = new ForwardIterator(this, read_options, cfd, sv);
iterators->push_back(
NewDBIterator(env_, *cfd->ioptions(), cfd->user_comparator(), iter,
kMaxSequenceNumber,
sv->mutable_cf_options.max_sequential_skip_in_iterations));
}
#endif
} else {
SequenceNumber latest_snapshot = versions_->LastSequence();
for (size_t i = 0; i < column_families.size(); ++i) {
auto* cfd = reinterpret_cast<ColumnFamilyHandleImpl*>(
column_families[i])->cfd();
SuperVersion* sv = cfd->GetReferencedSuperVersion(&mutex_);
auto snapshot =
read_options.snapshot != nullptr
? reinterpret_cast<const SnapshotImpl*>(
read_options.snapshot)->number_
: latest_snapshot;
ArenaWrappedDBIter* db_iter = NewArenaWrappedDbIterator(
env_, *cfd->ioptions(), cfd->user_comparator(), snapshot,
sv->mutable_cf_options.max_sequential_skip_in_iterations);
Iterator* internal_iter = NewInternalIterator(
read_options, cfd, sv, db_iter->GetArena());
db_iter->SetIterUnderDBIter(internal_iter);
iterators->push_back(db_iter);
}
}
return Status::OK();
}
Add a new mem-table representation based on cuckoo hash. Summary: = Major Changes = * Add a new mem-table representation, HashCuckooRep, which is based cuckoo hash. Cuckoo hash uses multiple hash functions. This allows each key to have multiple possible locations in the mem-table. - Put: When insert a key, it will try to find whether one of its possible locations is vacant and store the key. If none of its possible locations are available, then it will kick out a victim key and store at that location. The kicked-out victim key will then be stored at a vacant space of its possible locations or kick-out another victim. In this diff, the kick-out path (known as cuckoo-path) is found using BFS, which guarantees to be the shortest. - Get: Simply tries all possible locations of a key --- this guarantees worst-case constant time complexity. - Time complexity: O(1) for Get, and average O(1) for Put if the fullness of the mem-table is below 80%. - Default using two hash functions, the number of hash functions used by the cuckoo-hash may dynamically increase if it fails to find a short-enough kick-out path. - Currently, HashCuckooRep does not support iteration and snapshots, as our current main purpose of this is to optimize point access. = Minor Changes = * Add IsSnapshotSupported() to DB to indicate whether the current DB supports snapshots. If it returns false, then DB::GetSnapshot() will always return nullptr. Test Plan: Run existing tests. Will develop a test specifically for cuckoo hash in the next diff. Reviewers: sdong, haobo Reviewed By: sdong CC: leveldb, dhruba, igor Differential Revision: https://reviews.facebook.net/D16155
11 years ago
bool DBImpl::IsSnapshotSupported() const {
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (!cfd->mem()->IsSnapshotSupported()) {
return false;
}
}
return true;
}
const Snapshot* DBImpl::GetSnapshot() {
MutexLock l(&mutex_);
Add a new mem-table representation based on cuckoo hash. Summary: = Major Changes = * Add a new mem-table representation, HashCuckooRep, which is based cuckoo hash. Cuckoo hash uses multiple hash functions. This allows each key to have multiple possible locations in the mem-table. - Put: When insert a key, it will try to find whether one of its possible locations is vacant and store the key. If none of its possible locations are available, then it will kick out a victim key and store at that location. The kicked-out victim key will then be stored at a vacant space of its possible locations or kick-out another victim. In this diff, the kick-out path (known as cuckoo-path) is found using BFS, which guarantees to be the shortest. - Get: Simply tries all possible locations of a key --- this guarantees worst-case constant time complexity. - Time complexity: O(1) for Get, and average O(1) for Put if the fullness of the mem-table is below 80%. - Default using two hash functions, the number of hash functions used by the cuckoo-hash may dynamically increase if it fails to find a short-enough kick-out path. - Currently, HashCuckooRep does not support iteration and snapshots, as our current main purpose of this is to optimize point access. = Minor Changes = * Add IsSnapshotSupported() to DB to indicate whether the current DB supports snapshots. If it returns false, then DB::GetSnapshot() will always return nullptr. Test Plan: Run existing tests. Will develop a test specifically for cuckoo hash in the next diff. Reviewers: sdong, haobo Reviewed By: sdong CC: leveldb, dhruba, igor Differential Revision: https://reviews.facebook.net/D16155
11 years ago
// returns null if the underlying memtable does not support snapshot.
if (!IsSnapshotSupported()) return nullptr;
return snapshots_.New(versions_->LastSequence());
}
void DBImpl::ReleaseSnapshot(const Snapshot* s) {
MutexLock l(&mutex_);
snapshots_.Delete(reinterpret_cast<const SnapshotImpl*>(s));
}
// Convenience methods
Status DBImpl::Put(const WriteOptions& o, ColumnFamilyHandle* column_family,
const Slice& key, const Slice& val) {
return DB::Put(o, column_family, key, val);
}
Status DBImpl::Merge(const WriteOptions& o, ColumnFamilyHandle* column_family,
const Slice& key, const Slice& val) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
if (!cfh->cfd()->ioptions()->merge_operator) {
return Status::NotSupported("Provide a merge_operator when opening DB");
} else {
return DB::Merge(o, column_family, key, val);
}
}
Status DBImpl::Delete(const WriteOptions& write_options,
ColumnFamilyHandle* column_family, const Slice& key) {
return DB::Delete(write_options, column_family, key);
}
Status DBImpl::Write(const WriteOptions& write_options, WriteBatch* my_batch) {
if (my_batch == nullptr) {
return Status::Corruption("Batch is nullptr!");
}
PERF_TIMER_GUARD(write_pre_and_post_process_time);
WriteThread::Writer w(&mutex_);
w.batch = my_batch;
w.sync = write_options.sync;
w.disableWAL = write_options.disableWAL;
w.in_batch_group = false;
w.done = false;
w.timeout_hint_us = write_options.timeout_hint_us;
uint64_t expiration_time = 0;
bool has_timeout = false;
if (w.timeout_hint_us == 0) {
w.timeout_hint_us = WriteThread::kNoTimeOut;
} else {
expiration_time = env_->NowMicros() + w.timeout_hint_us;
has_timeout = true;
}
if (!write_options.disableWAL) {
RecordTick(stats_, WRITE_WITH_WAL);
default_cf_internal_stats_->AddDBStats(InternalStats::WRITE_WITH_WAL, 1);
}
WriteContext context;
mutex_.Lock();
Status status = write_thread_.EnterWriteThread(&w, expiration_time);
assert(status.ok() || status.IsTimedOut());
if (status.IsTimedOut()) {
mutex_.Unlock();
RecordTick(stats_, WRITE_TIMEDOUT);
return Status::TimedOut();
}
if (w.done) { // write was done by someone else
default_cf_internal_stats_->AddDBStats(InternalStats::WRITE_DONE_BY_OTHER,
1);
mutex_.Unlock();
RecordTick(stats_, WRITE_DONE_BY_OTHER);
return w.status;
}
RecordTick(stats_, WRITE_DONE_BY_SELF);
default_cf_internal_stats_->AddDBStats(InternalStats::WRITE_DONE_BY_SELF, 1);
// Once reaches this point, the current writer "w" will try to do its write
// job. It may also pick up some of the remaining writers in the "writers_"
// when it finds suitable, and finish them in the same write batch.
// This is how a write job could be done by the other writer.
assert(!single_column_family_mode_ ||
versions_->GetColumnFamilySet()->NumberOfColumnFamilies() == 1);
uint64_t max_total_wal_size = (db_options_.max_total_wal_size == 0)
? 4 * max_total_in_memory_state_
: db_options_.max_total_wal_size;
if (UNLIKELY(!single_column_family_mode_) &&
alive_log_files_.begin()->getting_flushed == false &&
total_log_size_ > max_total_wal_size) {
uint64_t flush_column_family_if_log_file = alive_log_files_.begin()->number;
alive_log_files_.begin()->getting_flushed = true;
Log(db_options_.info_log,
"Flushing all column families with data in WAL number %" PRIu64
". Total log size is %" PRIu64 " while max_total_wal_size is %" PRIu64,
flush_column_family_if_log_file, total_log_size_, max_total_wal_size);
// no need to refcount because drop is happening in write thread, so can't
// happen while we're in the write thread
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->GetLogNumber() <= flush_column_family_if_log_file) {
status = SetNewMemtableAndNewLogFile(cfd, &context);
if (!status.ok()) {
break;
}
cfd->imm()->FlushRequested();
}
}
MaybeScheduleFlushOrCompaction();
}
if (UNLIKELY(status.ok() && !bg_error_.ok())) {
status = bg_error_;
}
if (UNLIKELY(status.ok() && !flush_scheduler_.Empty())) {
status = ScheduleFlushes(&context);
}
if (UNLIKELY(status.ok()) &&
(write_controller_.IsStopped() || write_controller_.GetDelay() > 0)) {
status = DelayWrite(expiration_time);
}
if (UNLIKELY(status.ok() && has_timeout &&
env_->NowMicros() > expiration_time)) {
status = Status::TimedOut();
}
uint64_t last_sequence = versions_->LastSequence();
WriteThread::Writer* last_writer = &w;
if (status.ok()) {
autovector<WriteBatch*> write_batch_group;
write_thread_.BuildBatchGroup(&last_writer, &write_batch_group);
// Add to log and apply to memtable. We can release the lock
// during this phase since &w is currently responsible for logging
// and protects against concurrent loggers and concurrent writes
// into memtables
{
mutex_.Unlock();
WriteBatch* updates = nullptr;
if (write_batch_group.size() == 1) {
updates = write_batch_group[0];
} else {
updates = &tmp_batch_;
for (size_t i = 0; i < write_batch_group.size(); ++i) {
WriteBatchInternal::Append(updates, write_batch_group[i]);
}
}
const SequenceNumber current_sequence = last_sequence + 1;
WriteBatchInternal::SetSequence(updates, current_sequence);
int my_batch_count = WriteBatchInternal::Count(updates);
last_sequence += my_batch_count;
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
const uint64_t batch_size = WriteBatchInternal::ByteSize(updates);
// Record statistics
RecordTick(stats_, NUMBER_KEYS_WRITTEN, my_batch_count);
RecordTick(stats_, BYTES_WRITTEN, batch_size);
if (write_options.disableWAL) {
flush_on_destroy_ = true;
}
PERF_TIMER_STOP(write_pre_and_post_process_time);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
uint64_t log_size = 0;
if (!write_options.disableWAL) {
PERF_TIMER_GUARD(write_wal_time);
Add monitoring for universal compaction and add counters for compaction IO Summary: Adds these counters { WAL_FILE_SYNCED, "rocksdb.wal.synced" } number of writes that request a WAL sync { WAL_FILE_BYTES, "rocksdb.wal.bytes" }, number of bytes written to the WAL { WRITE_DONE_BY_SELF, "rocksdb.write.self" }, number of writes processed by the calling thread { WRITE_DONE_BY_OTHER, "rocksdb.write.other" }, number of writes not processed by the calling thread. Instead these were processed by the current holder of the write lock { WRITE_WITH_WAL, "rocksdb.write.wal" }, number of writes that request WAL logging { COMPACT_READ_BYTES, "rocksdb.compact.read.bytes" }, number of bytes read during compaction { COMPACT_WRITE_BYTES, "rocksdb.compact.write.bytes" }, number of bytes written during compaction Per-interval stats output was updated with WAL stats and correct stats for universal compaction including a correct value for write-amplification. It now looks like: Compactions Level Files Size(MB) Score Time(sec) Read(MB) Write(MB) Rn(MB) Rnp1(MB) Wnew(MB) RW-Amplify Read(MB/s) Write(MB/s) Rn Rnp1 Wnp1 NewW Count Ln-stall Stall-cnt -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 0 7 464 46.4 281 3411 3875 3411 0 3875 2.1 12.1 13.8 621 0 240 240 628 0.0 0 Uptime(secs): 310.8 total, 2.0 interval Writes cumulative: 9999999 total, 9999999 batches, 1.0 per batch, 1.22 ingest GB WAL cumulative: 9999999 WAL writes, 9999999 WAL syncs, 1.00 writes per sync, 1.22 GB written Compaction IO cumulative (GB): 1.22 new, 3.33 read, 3.78 write, 7.12 read+write Compaction IO cumulative (MB/sec): 4.0 new, 11.0 read, 12.5 write, 23.4 read+write Amplification cumulative: 4.1 write, 6.8 compaction Writes interval: 100000 total, 100000 batches, 1.0 per batch, 12.5 ingest MB WAL interval: 100000 WAL writes, 100000 WAL syncs, 1.00 writes per sync, 0.01 MB written Compaction IO interval (MB): 12.49 new, 14.98 read, 21.50 write, 36.48 read+write Compaction IO interval (MB/sec): 6.4 new, 7.6 read, 11.0 write, 18.6 read+write Amplification interval: 101.7 write, 102.9 compaction Stalls(secs): 142.924 level0_slowdown, 0.000 level0_numfiles, 0.805 memtable_compaction, 0.000 leveln_slowdown Stalls(count): 132461 level0_slowdown, 0 level0_numfiles, 3 memtable_compaction, 0 leveln_slowdown Task ID: #3329644, #3301695 Blame Rev: Test Plan: Revert Plan: Database Impact: Memcache Impact: Other Notes: EImportant: - begin *PUBLIC* platform impact section - Bugzilla: # - end platform impact - Reviewers: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D14583
11 years ago
Slice log_entry = WriteBatchInternal::Contents(updates);
status = log_->AddRecord(log_entry);
total_log_size_ += log_entry.size();
alive_log_files_.back().AddSize(log_entry.size());
log_empty_ = false;
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
log_size = log_entry.size();
RecordTick(stats_, WAL_FILE_BYTES, log_size);
if (status.ok() && write_options.sync) {
RecordTick(stats_, WAL_FILE_SYNCED);
StopWatch sw(env_, stats_, WAL_FILE_SYNC_MICROS);
if (db_options_.use_fsync) {
status = log_->file()->Fsync();
} else {
status = log_->file()->Sync();
}
}
}
if (status.ok()) {
PERF_TIMER_GUARD(write_memtable_time);
status = WriteBatchInternal::InsertInto(
updates, column_family_memtables_.get(),
write_options.ignore_missing_column_families, 0, this, false);
// A non-OK status here indicates iteration failure (either in-memory
// writebatch corruption (very bad), or the client specified invalid
// column family). This will later on trigger bg_error_.
//
// Note that existing logic was not sound. Any partial failure writing
// into the memtable would result in a state that some write ops might
// have succeeded in memtable but Status reports error for all writes.
SetTickerCount(stats_, SEQUENCE_NUMBER, last_sequence);
}
PERF_TIMER_START(write_pre_and_post_process_time);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
if (updates == &tmp_batch_) {
tmp_batch_.Clear();
}
mutex_.Lock();
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
// internal stats
default_cf_internal_stats_->AddDBStats(
InternalStats::BYTES_WRITTEN, batch_size);
default_cf_internal_stats_->AddDBStats(InternalStats::NUMBER_KEYS_WRITTEN,
my_batch_count);
if (!write_options.disableWAL) {
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
10 years ago
default_cf_internal_stats_->AddDBStats(
InternalStats::WAL_FILE_SYNCED, 1);
default_cf_internal_stats_->AddDBStats(
InternalStats::WAL_FILE_BYTES, log_size);
}
if (status.ok()) {
versions_->SetLastSequence(last_sequence);
}
}
}
if (db_options_.paranoid_checks && !status.ok() &&
!status.IsTimedOut() && bg_error_.ok()) {
bg_error_ = status; // stop compaction & fail any further writes
}
write_thread_.ExitWriteThread(&w, last_writer, status);
mutex_.Unlock();
if (status.IsTimedOut()) {
RecordTick(stats_, WRITE_TIMEDOUT);
}
return status;
}
// REQUIRES: mutex_ is held
// REQUIRES: this thread is currently at the front of the writer queue
Status DBImpl::DelayWrite(uint64_t expiration_time) {
StopWatch sw(env_, stats_, WRITE_STALL);
bool has_timeout = (expiration_time > 0);
auto delay = write_controller_.GetDelay();
if (write_controller_.IsStopped() == false && delay > 0) {
mutex_.Unlock();
env_->SleepForMicroseconds(delay);
mutex_.Lock();
}
while (bg_error_.ok() && write_controller_.IsStopped()) {
if (has_timeout) {
bg_cv_.TimedWait(expiration_time);
if (env_->NowMicros() > expiration_time) {
return Status::TimedOut();
}
} else {
bg_cv_.Wait();
}
}
return bg_error_;
}
Status DBImpl::ScheduleFlushes(WriteContext* context) {
bool schedule_bg_work = false;
ColumnFamilyData* cfd;
while ((cfd = flush_scheduler_.GetNextColumnFamily()) != nullptr) {
schedule_bg_work = true;
auto status = SetNewMemtableAndNewLogFile(cfd, context);
if (cfd->Unref()) {
delete cfd;
}
if (!status.ok()) {
return status;
}
}
if (schedule_bg_work) {
MaybeScheduleFlushOrCompaction();
}
return Status::OK();
}
// REQUIRES: mutex_ is held
// REQUIRES: this thread is currently at the front of the writer queue
Status DBImpl::SetNewMemtableAndNewLogFile(ColumnFamilyData* cfd,
WriteContext* context) {
mutex_.AssertHeld();
unique_ptr<WritableFile> lfile;
log::Writer* new_log = nullptr;
MemTable* new_mem = nullptr;
// Attempt to switch to a new memtable and trigger flush of old.
// Do this without holding the dbmutex lock.
assert(versions_->PrevLogNumber() == 0);
bool creating_new_log = !log_empty_;
uint64_t new_log_number =
creating_new_log ? versions_->NewFileNumber() : logfile_number_;
SuperVersion* new_superversion = nullptr;
const MutableCFOptions mutable_cf_options = *cfd->GetLatestMutableCFOptions();
mutex_.Unlock();
Status s;
{
if (creating_new_log) {
s = env_->NewWritableFile(
LogFileName(db_options_.wal_dir, new_log_number),
&lfile, env_->OptimizeForLogWrite(env_options_));
if (s.ok()) {
// Our final size should be less than write_buffer_size
// (compression, etc) but err on the side of caution.
lfile->SetPreallocationBlockSize(
1.1 * mutable_cf_options.write_buffer_size);
new_log = new log::Writer(std::move(lfile));
}
}
if (s.ok()) {
new_mem = new MemTable(cfd->internal_comparator(), *cfd->ioptions(),
mutable_cf_options);
new_superversion = new SuperVersion();
}
}
Log(db_options_.info_log,
"[%s] New memtable created with log file: #%" PRIu64 "\n",
cfd->GetName().c_str(), new_log_number);
mutex_.Lock();
if (!s.ok()) {
// how do we fail if we're not creating new log?
assert(creating_new_log);
// Avoid chewing through file number space in a tight loop.
versions_->ReuseLogFileNumber(new_log_number);
assert(!new_mem);
assert(!new_log);
return s;
}
if (creating_new_log) {
logfile_number_ = new_log_number;
assert(new_log != nullptr);
context->logs_to_free_.push_back(log_.release());
log_.reset(new_log);
log_empty_ = true;
alive_log_files_.push_back(LogFileNumberSize(logfile_number_));
for (auto cfd : *versions_->GetColumnFamilySet()) {
// all this is just optimization to delete logs that
// are no longer needed -- if CF is empty, that means it
// doesn't need that particular log to stay alive, so we just
// advance the log number. no need to persist this in the manifest
if (cfd->mem()->GetFirstSequenceNumber() == 0 &&
cfd->imm()->size() == 0) {
cfd->SetLogNumber(logfile_number_);
}
}
}
cfd->mem()->SetNextLogNumber(logfile_number_);
cfd->imm()->Add(cfd->mem());
new_mem->Ref();
cfd->SetMemtable(new_mem);
context->superversions_to_free_.push_back(
InstallSuperVersion(cfd, new_superversion, mutable_cf_options));
return s;
}
#ifndef ROCKSDB_LITE
Status DBImpl::GetPropertiesOfAllTables(ColumnFamilyHandle* column_family,
TablePropertiesCollection* props) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
// Increment the ref count
mutex_.Lock();
auto version = cfd->current();
version->Ref();
mutex_.Unlock();
auto s = version->GetPropertiesOfAllTables(props);
// Decrement the ref count
mutex_.Lock();
version->Unref();
mutex_.Unlock();
return s;
}
#endif // ROCKSDB_LITE
[RocksDB] BackupableDB Summary: In this diff I present you BackupableDB v1. You can easily use it to backup your DB and it will do incremental snapshots for you. Let's first describe how you would use BackupableDB. It's inheriting StackableDB interface so you can easily construct it with your DB object -- it will add a method RollTheSnapshot() to the DB object. When you call RollTheSnapshot(), current snapshot of the DB will be stored in the backup dir. To restore, you can just call RestoreDBFromBackup() on a BackupableDB (which is a static method) and it will restore all files from the backup dir. In the next version, it will even support automatic backuping every X minutes. There are multiple things you can configure: 1. backup_env and db_env can be different, which is awesome because then you can easily backup to HDFS or wherever you feel like. 2. sync - if true, it *guarantees* backup consistency on machine reboot 3. number of snapshots to keep - this will keep last N snapshots around if you want, for some reason, be able to restore from an earlier snapshot. All the backuping is done in incremental fashion - if we already have 00010.sst, we will not copy it again. *IMPORTANT* -- This is based on assumption that 00010.sst never changes - two files named 00010.sst from the same DB will always be exactly the same. Is this true? I always copy manifest, current and log files. 4. You can decide if you want to flush the memtables before you backup, or you're fine with backing up the log files -- either way, you get a complete and consistent view of the database at a time of backup. 5. More things you can find in BackupableDBOptions Here is the directory structure I use: backup_dir/CURRENT_SNAPSHOT - just 4 bytes holding the latest snapshot 0, 1, 2, ... - files containing serialized version of each snapshot - containing a list of files files/*.sst - sst files shared between snapshots - if one snapshot references 00010.sst and another one needs to backup it from the DB, it will just reference the same file files/ 0/, 1/, 2/, ... - snapshot directories containing private snapshot files - current, manifest and log files All the files are ref counted and deleted immediatelly when they get out of scope. Some other stuff in this diff: 1. Added GetEnv() method to the DB. Discussed with @haobo and we agreed that it seems right thing to do. 2. Fixed StackableDB interface. The way it was set up before, I was not able to implement BackupableDB. Test Plan: I have a unittest, but please don't look at this yet. I just hacked it up to help me with debugging. I will write a lot of good tests and update the diff. Also, `make asan_check` Reviewers: dhruba, haobo, emayanke Reviewed By: dhruba CC: leveldb, haobo Differential Revision: https://reviews.facebook.net/D14295
11 years ago
const std::string& DBImpl::GetName() const {
return dbname_;
}
Env* DBImpl::GetEnv() const {
return env_;
}
const Options& DBImpl::GetOptions(ColumnFamilyHandle* column_family) const {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
return *cfh->cfd()->options();
}
bool DBImpl::GetProperty(ColumnFamilyHandle* column_family,
const Slice& property, std::string* value) {
bool is_int_property = false;
bool need_out_of_mutex = false;
DBPropertyType property_type =
GetPropertyType(property, &is_int_property, &need_out_of_mutex);
value->clear();
if (is_int_property) {
uint64_t int_value;
bool ret_value = GetIntPropertyInternal(column_family, property_type,
need_out_of_mutex, &int_value);
if (ret_value) {
*value = std::to_string(int_value);
}
return ret_value;
} else {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
MutexLock l(&mutex_);
return cfd->internal_stats()->GetStringProperty(property_type, property,
value);
}
}
bool DBImpl::GetIntProperty(ColumnFamilyHandle* column_family,
const Slice& property, uint64_t* value) {
bool is_int_property = false;
bool need_out_of_mutex = false;
DBPropertyType property_type =
GetPropertyType(property, &is_int_property, &need_out_of_mutex);
if (!is_int_property) {
return false;
}
return GetIntPropertyInternal(column_family, property_type, need_out_of_mutex,
value);
}
bool DBImpl::GetIntPropertyInternal(ColumnFamilyHandle* column_family,
DBPropertyType property_type,
bool need_out_of_mutex, uint64_t* value) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
if (!need_out_of_mutex) {
MutexLock l(&mutex_);
return cfd->internal_stats()->GetIntProperty(property_type, value, this);
} else {
SuperVersion* sv = GetAndRefSuperVersion(cfd);
bool ret = cfd->internal_stats()->GetIntPropertyOutOfMutex(
property_type, sv->current, value);
ReturnAndCleanupSuperVersion(cfd, sv);
return ret;
}
}
SuperVersion* DBImpl::GetAndRefSuperVersion(ColumnFamilyData* cfd) {
// TODO(ljin): consider using GetReferencedSuperVersion() directly
return cfd->GetThreadLocalSuperVersion(&mutex_);
}
void DBImpl::ReturnAndCleanupSuperVersion(ColumnFamilyData* cfd,
SuperVersion* sv) {
bool unref_sv = !cfd->ReturnThreadLocalSuperVersion(sv);
if (unref_sv) {
// Release SuperVersion
if (sv->Unref()) {
{
MutexLock l(&mutex_);
sv->Cleanup();
}
delete sv;
RecordTick(stats_, NUMBER_SUPERVERSION_CLEANUPS);
}
RecordTick(stats_, NUMBER_SUPERVERSION_RELEASES);
}
}
void DBImpl::GetApproximateSizes(ColumnFamilyHandle* column_family,
const Range* range, int n, uint64_t* sizes) {
// TODO(opt): better implementation
Version* v;
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
{
MutexLock l(&mutex_);
v = cfd->current();
v->Ref();
}
for (int i = 0; i < n; i++) {
// Convert user_key into a corresponding internal key.
InternalKey k1(range[i].start, kMaxSequenceNumber, kValueTypeForSeek);
InternalKey k2(range[i].limit, kMaxSequenceNumber, kValueTypeForSeek);
uint64_t start = versions_->ApproximateOffsetOf(v, k1);
uint64_t limit = versions_->ApproximateOffsetOf(v, k2);
sizes[i] = (limit >= start ? limit - start : 0);
}
{
MutexLock l(&mutex_);
v->Unref();
}
}
#ifndef ROCKSDB_LITE
Status DBImpl::GetUpdatesSince(
SequenceNumber seq, unique_ptr<TransactionLogIterator>* iter,
const TransactionLogIterator::ReadOptions& read_options) {
RecordTick(stats_, GET_UPDATES_SINCE_CALLS);
if (seq > versions_->LastSequence()) {
return Status::NotFound("Requested sequence not yet written in the db");
}
// Get all sorted Wal Files.
// Do binary search and open files and find the seq number.
std::unique_ptr<VectorLogPtr> wal_files(new VectorLogPtr);
Status s = GetSortedWalFiles(*wal_files);
if (!s.ok()) {
return s;
}
s = RetainProbableWalFiles(*wal_files, seq);
if (!s.ok()) {
return s;
}
iter->reset(new TransactionLogIteratorImpl(db_options_.wal_dir, &db_options_,
read_options, env_options_,
seq, std::move(wal_files), this));
return (*iter)->status();
}
Status DBImpl::DeleteFile(std::string name) {
uint64_t number;
FileType type;
WalFileType log_type;
if (!ParseFileName(name, &number, &type, &log_type) ||
(type != kTableFile && type != kLogFile)) {
Log(db_options_.info_log, "DeleteFile %s failed.\n", name.c_str());
return Status::InvalidArgument("Invalid file name");
}
Status status;
if (type == kLogFile) {
// Only allow deleting archived log files
if (log_type != kArchivedLogFile) {
Log(db_options_.info_log, "DeleteFile %s failed - not archived log.\n",
name.c_str());
return Status::NotSupported("Delete only supported for archived logs");
}
status = env_->DeleteFile(db_options_.wal_dir + "/" + name.c_str());
if (!status.ok()) {
Log(db_options_.info_log, "DeleteFile %s failed -- %s.\n",
name.c_str(), status.ToString().c_str());
}
return status;
}
int level;
FileMetaData* metadata;
ColumnFamilyData* cfd;
VersionEdit edit;
JobContext job_context(true);
{
MutexLock l(&mutex_);
status = versions_->GetMetadataForFile(number, &level, &metadata, &cfd);
if (!status.ok()) {
Log(db_options_.info_log, "DeleteFile %s failed. File not found\n",
name.c_str());
return Status::InvalidArgument("File not found");
}
assert(level < cfd->NumberLevels());
// If the file is being compacted no need to delete.
if (metadata->being_compacted) {
Log(db_options_.info_log,
"DeleteFile %s Skipped. File about to be compacted\n", name.c_str());
return Status::OK();
}
// Only the files in the last level can be deleted externally.
// This is to make sure that any deletion tombstones are not
// lost. Check that the level passed is the last level.
auto* vstoreage = cfd->current()->GetStorageInfo();
for (int i = level + 1; i < cfd->NumberLevels(); i++) {
if (vstoreage->NumLevelFiles(i) != 0) {
Log(db_options_.info_log,
"DeleteFile %s FAILED. File not in last level\n", name.c_str());
return Status::InvalidArgument("File not in last level");
}
}
// if level == 0, it has to be the oldest file
if (level == 0 && vstoreage->files_[0].back()->fd.GetNumber() != number) {
return Status::InvalidArgument("File in level 0, but not oldest");
}
edit.SetColumnFamily(cfd->GetID());
edit.DeleteFile(level, number);
status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
&edit, &mutex_, db_directory_.get());
if (status.ok()) {
InstallSuperVersionBackground(cfd, &job_context,
*cfd->GetLatestMutableCFOptions());
}
FindObsoleteFiles(&job_context, false);
} // lock released here
LogFlush(db_options_.info_log);
// remove files outside the db-lock
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
{
MutexLock l(&mutex_);
// schedule flush if file deletion means we freed the space for flushes to
// continue
MaybeScheduleFlushOrCompaction();
}
return status;
}
void DBImpl::GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {
MutexLock l(&mutex_);
versions_->GetLiveFilesMetaData(metadata);
}
#endif // ROCKSDB_LITE
Status DBImpl::CheckConsistency() {
mutex_.AssertHeld();
std::vector<LiveFileMetaData> metadata;
versions_->GetLiveFilesMetaData(&metadata);
std::string corruption_messages;
for (const auto& md : metadata) {
std::string file_path = md.db_path + "/" + md.name;
uint64_t fsize = 0;
Status s = env_->GetFileSize(file_path, &fsize);
if (!s.ok()) {
corruption_messages +=
"Can't access " + md.name + ": " + s.ToString() + "\n";
} else if (fsize != md.size) {
corruption_messages += "Sst file size mismatch: " + file_path +
". Size recorded in manifest " +
std::to_string(md.size) + ", actual size " +
std::to_string(fsize) + "\n";
}
}
if (corruption_messages.size() == 0) {
return Status::OK();
} else {
return Status::Corruption(corruption_messages);
}
}
Status DBImpl::GetDbIdentity(std::string& identity) {
std::string idfilename = IdentityFileName(dbname_);
unique_ptr<SequentialFile> idfile;
const EnvOptions soptions;
Status s = env_->NewSequentialFile(idfilename, &idfile, soptions);
if (!s.ok()) {
return s;
}
uint64_t file_size;
s = env_->GetFileSize(idfilename, &file_size);
if (!s.ok()) {
return s;
}
char buffer[file_size];
Slice id;
s = idfile->Read(file_size, &id, buffer);
if (!s.ok()) {
return s;
}
identity.assign(id.ToString());
// If last character is '\n' remove it from identity
if (identity.size() > 0 && identity.back() == '\n') {
identity.pop_back();
}
return s;
}
// Default implementations of convenience methods that subclasses of DB
// can call if they wish
Status DB::Put(const WriteOptions& opt, ColumnFamilyHandle* column_family,
const Slice& key, const Slice& value) {
// Pre-allocate size of write batch conservatively.
// 8 bytes are taken by header, 4 bytes for count, 1 byte for type,
// and we allocate 11 extra bytes for key length, as well as value length.
WriteBatch batch(key.size() + value.size() + 24);
batch.Put(column_family, key, value);
return Write(opt, &batch);
}
Status DB::Delete(const WriteOptions& opt, ColumnFamilyHandle* column_family,
const Slice& key) {
WriteBatch batch;
batch.Delete(column_family, key);
return Write(opt, &batch);
}
Status DB::Merge(const WriteOptions& opt, ColumnFamilyHandle* column_family,
const Slice& key, const Slice& value) {
WriteBatch batch;
batch.Merge(column_family, key, value);
return Write(opt, &batch);
}
// Default implementation -- returns not supported status
Status DB::CreateColumnFamily(const ColumnFamilyOptions& cf_options,
const std::string& column_family_name,
ColumnFamilyHandle** handle) {
return Status::NotSupported("");
}
Status DB::DropColumnFamily(ColumnFamilyHandle* column_family) {
return Status::NotSupported("");
}
DB::~DB() { }
Status DB::Open(const Options& options, const std::string& dbname, DB** dbptr) {
DBOptions db_options(options);
ColumnFamilyOptions cf_options(options);
std::vector<ColumnFamilyDescriptor> column_families;
column_families.push_back(
ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
std::vector<ColumnFamilyHandle*> handles;
Status s = DB::Open(db_options, dbname, column_families, &handles, dbptr);
if (s.ok()) {
assert(handles.size() == 1);
// i can delete the handle since DBImpl is always holding a reference to
// default column family
delete handles[0];
}
return s;
}
Status DB::Open(const DBOptions& db_options, const std::string& dbname,
const std::vector<ColumnFamilyDescriptor>& column_families,
std::vector<ColumnFamilyHandle*>* handles, DB** dbptr) {
Status s = SanitizeOptionsByTable(db_options, column_families);
if (!s.ok()) {
return s;
}
if (db_options.db_paths.size() > 1) {
for (auto& cfd : column_families) {
if (cfd.options.compaction_style != kCompactionStyleUniversal) {
return Status::NotSupported(
"More than one DB paths are only supported in "
"universal compaction style. ");
}
}
if (db_options.db_paths.size() > 4) {
return Status::NotSupported(
"More than four DB paths are not supported yet. ");
}
}
*dbptr = nullptr;
handles->clear();
size_t max_write_buffer_size = 0;
for (auto cf : column_families) {
max_write_buffer_size =
std::max(max_write_buffer_size, cf.options.write_buffer_size);
}
DBImpl* impl = new DBImpl(db_options, dbname);
s = impl->env_->CreateDirIfMissing(impl->db_options_.wal_dir);
if (s.ok()) {
for (auto db_path : impl->db_options_.db_paths) {
s = impl->env_->CreateDirIfMissing(db_path.path);
if (!s.ok()) {
break;
}
}
}
if (!s.ok()) {
delete impl;
return s;
}
s = impl->CreateArchivalDirectory();
if (!s.ok()) {
delete impl;
return s;
}
impl->mutex_.Lock();
// Handles create_if_missing, error_if_exists
s = impl->Recover(column_families);
if (s.ok()) {
uint64_t new_log_number = impl->versions_->NewFileNumber();
unique_ptr<WritableFile> lfile;
EnvOptions soptions(db_options);
s = impl->db_options_.env->NewWritableFile(
LogFileName(impl->db_options_.wal_dir, new_log_number), &lfile,
impl->db_options_.env->OptimizeForLogWrite(soptions));
if (s.ok()) {
lfile->SetPreallocationBlockSize(1.1 * max_write_buffer_size);
impl->logfile_number_ = new_log_number;
impl->log_.reset(new log::Writer(std::move(lfile)));
// set column family handles
for (auto cf : column_families) {
auto cfd =
impl->versions_->GetColumnFamilySet()->GetColumnFamily(cf.name);
if (cfd != nullptr) {
handles->push_back(
new ColumnFamilyHandleImpl(cfd, impl, &impl->mutex_));
} else {
if (db_options.create_missing_column_families) {
// missing column family, create it
ColumnFamilyHandle* handle;
impl->mutex_.Unlock();
s = impl->CreateColumnFamily(cf.options, cf.name, &handle);
impl->mutex_.Lock();
if (s.ok()) {
handles->push_back(handle);
} else {
break;
}
} else {
s = Status::InvalidArgument("Column family not found: ", cf.name);
break;
}
}
}
}
if (s.ok()) {
for (auto cfd : *impl->versions_->GetColumnFamilySet()) {
delete impl->InstallSuperVersion(
cfd, nullptr, *cfd->GetLatestMutableCFOptions());
}
impl->alive_log_files_.push_back(
DBImpl::LogFileNumberSize(impl->logfile_number_));
impl->DeleteObsoleteFiles();
impl->MaybeScheduleFlushOrCompaction();
s = impl->db_directory_->Fsync();
}
}
if (s.ok()) {
for (auto cfd : *impl->versions_->GetColumnFamilySet()) {
if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
auto* vstorage = cfd->current()->GetStorageInfo();
for (int i = 1; i < vstorage->NumberLevels(); ++i) {
int num_files = vstorage->NumLevelFiles(i);
if (num_files > 0) {
s = Status::InvalidArgument(
"Not all files are at level 0. Cannot "
"open with universal or FIFO compaction style.");
break;
}
}
}
if (cfd->ioptions()->merge_operator != nullptr &&
Add a new mem-table representation based on cuckoo hash. Summary: = Major Changes = * Add a new mem-table representation, HashCuckooRep, which is based cuckoo hash. Cuckoo hash uses multiple hash functions. This allows each key to have multiple possible locations in the mem-table. - Put: When insert a key, it will try to find whether one of its possible locations is vacant and store the key. If none of its possible locations are available, then it will kick out a victim key and store at that location. The kicked-out victim key will then be stored at a vacant space of its possible locations or kick-out another victim. In this diff, the kick-out path (known as cuckoo-path) is found using BFS, which guarantees to be the shortest. - Get: Simply tries all possible locations of a key --- this guarantees worst-case constant time complexity. - Time complexity: O(1) for Get, and average O(1) for Put if the fullness of the mem-table is below 80%. - Default using two hash functions, the number of hash functions used by the cuckoo-hash may dynamically increase if it fails to find a short-enough kick-out path. - Currently, HashCuckooRep does not support iteration and snapshots, as our current main purpose of this is to optimize point access. = Minor Changes = * Add IsSnapshotSupported() to DB to indicate whether the current DB supports snapshots. If it returns false, then DB::GetSnapshot() will always return nullptr. Test Plan: Run existing tests. Will develop a test specifically for cuckoo hash in the next diff. Reviewers: sdong, haobo Reviewed By: sdong CC: leveldb, dhruba, igor Differential Revision: https://reviews.facebook.net/D16155
11 years ago
!cfd->mem()->IsMergeOperatorSupported()) {
s = Status::InvalidArgument(
"The memtable of column family %s does not support merge operator "
"its options.merge_operator is non-null", cfd->GetName().c_str());
}
if (!s.ok()) {
break;
}
}
}
impl->mutex_.Unlock();
if (s.ok()) {
impl->opened_successfully_ = true;
*dbptr = impl;
} else {
for (auto h : *handles) {
delete h;
}
handles->clear();
delete impl;
}
return s;
}
Status DB::ListColumnFamilies(const DBOptions& db_options,
const std::string& name,
std::vector<std::string>* column_families) {
return VersionSet::ListColumnFamilies(column_families, name, db_options.env);
}
Snapshot::~Snapshot() {
}
Status DestroyDB(const std::string& dbname, const Options& options) {
const InternalKeyComparator comparator(options.comparator);
const Options& soptions(SanitizeOptions(dbname, &comparator, options));
Env* env = soptions.env;
std::vector<std::string> filenames;
std::vector<std::string> archiveFiles;
std::string archivedir = ArchivalDirectory(dbname);
// Ignore error in case directory does not exist
env->GetChildren(dbname, &filenames);
if (dbname != soptions.wal_dir) {
std::vector<std::string> logfilenames;
env->GetChildren(soptions.wal_dir, &logfilenames);
filenames.insert(filenames.end(), logfilenames.begin(), logfilenames.end());
archivedir = ArchivalDirectory(soptions.wal_dir);
}
if (filenames.empty()) {
return Status::OK();
}
FileLock* lock;
const std::string lockname = LockFileName(dbname);
Status result = env->LockFile(lockname, &lock);
if (result.ok()) {
uint64_t number;
FileType type;
InfoLogPrefix info_log_prefix(!options.db_log_dir.empty(), dbname);
for (size_t i = 0; i < filenames.size(); i++) {
if (ParseFileName(filenames[i], &number, info_log_prefix.prefix, &type) &&
type != kDBLockFile) { // Lock file will be deleted at end
Status del;
if (type == kMetaDatabase) {
del = DestroyDB(dbname + "/" + filenames[i], options);
} else if (type == kLogFile) {
del = env->DeleteFile(soptions.wal_dir + "/" + filenames[i]);
} else {
del = env->DeleteFile(dbname + "/" + filenames[i]);
}
if (result.ok() && !del.ok()) {
result = del;
}
}
}
for (auto& db_path : options.db_paths) {
env->GetChildren(db_path.path, &filenames);
uint64_t number;
FileType type;
for (size_t i = 0; i < filenames.size(); i++) {
if (ParseFileName(filenames[i], &number, &type) &&
type == kTableFile) { // Lock file will be deleted at end
Status del = env->DeleteFile(db_path.path + "/" + filenames[i]);
if (result.ok() && !del.ok()) {
result = del;
}
}
}
}
env->GetChildren(archivedir, &archiveFiles);
// Delete archival files.
for (size_t i = 0; i < archiveFiles.size(); ++i) {
if (ParseFileName(archiveFiles[i], &number, &type) &&
type == kLogFile) {
Status del = env->DeleteFile(archivedir + "/" + archiveFiles[i]);
if (result.ok() && !del.ok()) {
result = del;
}
}
}
// ignore case where no archival directory is present.
env->DeleteDir(archivedir);
env->UnlockFile(lock); // Ignore error since state is already gone
env->DeleteFile(lockname);
env->DeleteDir(dbname); // Ignore error in case dir contains other files
env->DeleteDir(soptions.wal_dir);
}
return result;
}
//
// A global method that can dump out the build version
void DumpRocksDBBuildVersion(Logger * log) {
#if !defined(IOS_CROSS_COMPILE)
// if we compile with Xcode, we don't run build_detect_vesion, so we don't
// generate util/build_version.cc
Log(log, "RocksDB version: %d.%d.%d\n", ROCKSDB_MAJOR, ROCKSDB_MINOR,
ROCKSDB_PATCH);
Log(log, "Git sha %s", rocksdb_build_git_sha);
Log(log, "Compile time %s %s",
rocksdb_build_compile_time, rocksdb_build_compile_date);
#endif
}
} // namespace rocksdb